id sid tid token lemma pos fcis-3177 1 1 frontiers frontier NOUN fcis-3177 1 2 in in ADP fcis-3177 1 3 computing computing NOUN fcis-3177 1 4 and and CCONJ fcis-3177 1 5 intelligent intelligent ADJ fcis-3177 1 6 systems system NOUN fcis-3177 1 7 issn issn VERB fcis-3177 1 8 : : PUNCT fcis-3177 1 9 2832 2832 NUM fcis-3177 1 10 - - SYM fcis-3177 1 11 6024 6024 NUM fcis-3177 1 12 | | NOUN fcis-3177 1 13 vol vol NOUN fcis-3177 1 14 . . PROPN fcis-3177 2 1 2 2 NUM fcis-3177 2 2 , , PUNCT fcis-3177 2 3 no no INTJ fcis-3177 2 4 . . NOUN fcis-3177 2 5 1 1 NUM fcis-3177 2 6 , , PUNCT fcis-3177 2 7 2022 2022 NUM fcis-3177 2 8 110 110 NUM fcis-3177 2 9 a a DET fcis-3177 2 10 survey survey NOUN fcis-3177 2 11 of of ADP fcis-3177 2 12 few few ADJ fcis-3177 2 13 - - PUNCT fcis-3177 2 14 shot shot NOUN fcis-3177 2 15 learning learn VERB fcis-3177 2 16 research research NOUN fcis-3177 2 17 based base VERB fcis-3177 2 18 on on ADP fcis-3177 2 19 deep deep ADJ fcis-3177 2 20 neural neural ADJ fcis-3177 2 21 network network NOUN fcis-3177 2 22 pengjin pengjin PROPN fcis-3177 2 23 wu wu PROPN fcis-3177 2 24 department department PROPN fcis-3177 2 25 of of ADP fcis-3177 2 26 electrical electrical ADJ fcis-3177 2 27 and and CCONJ fcis-3177 2 28 electronic electronic ADJ fcis-3177 2 29 engineering engineering NOUN fcis-3177 2 30 , , PUNCT fcis-3177 2 31 university university NOUN fcis-3177 2 32 of of ADP fcis-3177 2 33 surrey surrey PROPN fcis-3177 2 34 , , PUNCT fcis-3177 2 35 guildford guildford PROPN fcis-3177 2 36 gu2 gu2 PROPN fcis-3177 2 37 7xh 7xh NOUN fcis-3177 2 38 , , PUNCT fcis-3177 2 39 uk uk PROPN fcis-3177 2 40 abstract abstract NOUN fcis-3177 2 41 : : PUNCT fcis-3177 2 42 with with ADP fcis-3177 2 43 the the DET fcis-3177 2 44 successful successful ADJ fcis-3177 2 45 development development NOUN fcis-3177 2 46 of of ADP fcis-3177 2 47 deep deep ADJ fcis-3177 2 48 learning learning NOUN fcis-3177 2 49 techniques technique NOUN fcis-3177 2 50 in in ADP fcis-3177 2 51 recent recent ADJ fcis-3177 2 52 years year NOUN fcis-3177 2 53 , , PUNCT fcis-3177 2 54 deep deep ADJ fcis-3177 2 55 neural neural ADJ fcis-3177 2 56 networks network NOUN fcis-3177 2 57 have have AUX fcis-3177 2 58 achieved achieve VERB fcis-3177 2 59 excellent excellent ADJ fcis-3177 2 60 results result NOUN fcis-3177 2 61 in in ADP fcis-3177 2 62 both both PRON fcis-3177 2 63 computer computer NOUN fcis-3177 2 64 vision vision NOUN fcis-3177 2 65 and and CCONJ fcis-3177 2 66 natural natural ADJ fcis-3177 2 67 language language NOUN fcis-3177 2 68 processing processing NOUN fcis-3177 2 69 by by ADP fcis-3177 2 70 relying rely VERB fcis-3177 2 71 on on ADP fcis-3177 2 72 large large ADJ fcis-3177 2 73 - - PUNCT fcis-3177 2 74 scale scale NOUN fcis-3177 2 75 datasets dataset NOUN fcis-3177 2 76 but but CCONJ fcis-3177 2 77 still still ADV fcis-3177 2 78 face face VERB fcis-3177 2 79 significant significant ADJ fcis-3177 2 80 challenges challenge NOUN fcis-3177 2 81 in in ADP fcis-3177 2 82 solving solve VERB fcis-3177 2 83 the the DET fcis-3177 2 84 problem problem NOUN fcis-3177 2 85 of of ADP fcis-3177 2 86 learning learn VERB fcis-3177 2 87 from from ADP fcis-3177 2 88 few few ADJ fcis-3177 2 89 - - PUNCT fcis-3177 2 90 shot shot NOUN fcis-3177 2 91 . . PUNCT fcis-3177 3 1 inspired inspire VERB fcis-3177 3 2 by by ADP fcis-3177 3 3 the the DET fcis-3177 3 4 ability ability NOUN fcis-3177 3 5 of of ADP fcis-3177 3 6 humans human NOUN fcis-3177 3 7 to to PART fcis-3177 3 8 learn learn VERB fcis-3177 3 9 to to PART fcis-3177 3 10 recognize recognize VERB fcis-3177 3 11 objects object NOUN fcis-3177 3 12 as as ADP fcis-3177 3 13 a a DET fcis-3177 3 14 way way NOUN fcis-3177 3 15 to to PART fcis-3177 3 16 simulate simulate VERB fcis-3177 3 17 the the DET fcis-3177 3 18 cognitive cognitive ADJ fcis-3177 3 19 process process NOUN fcis-3177 3 20 of of ADP fcis-3177 3 21 learning learn VERB fcis-3177 3 22 from from ADP fcis-3177 3 23 a a DET fcis-3177 3 24 small small ADJ fcis-3177 3 25 sample sample NOUN fcis-3177 3 26 size size NOUN fcis-3177 3 27 , , PUNCT fcis-3177 3 28 few few ADJ fcis-3177 3 29 - - PUNCT fcis-3177 3 30 shot shot NOUN fcis-3177 3 31 learning learning NOUN fcis-3177 3 32 is be AUX fcis-3177 3 33 a a DET fcis-3177 3 34 hot hot ADJ fcis-3177 3 35 topic topic NOUN fcis-3177 3 36 of of ADP fcis-3177 3 37 research research NOUN fcis-3177 3 38 in in ADP fcis-3177 3 39 deep deep ADJ fcis-3177 3 40 neural neural ADJ fcis-3177 3 41 networks network NOUN fcis-3177 3 42 today today NOUN fcis-3177 3 43 . . PUNCT fcis-3177 4 1 it it PRON fcis-3177 4 2 is be AUX fcis-3177 4 3 also also ADV fcis-3177 4 4 a a DET fcis-3177 4 5 significant significant ADJ fcis-3177 4 6 and and CCONJ fcis-3177 4 7 challenging challenging ADJ fcis-3177 4 8 problem problem NOUN fcis-3177 4 9 . . PUNCT fcis-3177 5 1 this this DET fcis-3177 5 2 paper paper NOUN fcis-3177 5 3 first first ADV fcis-3177 5 4 introduces introduce VERB fcis-3177 5 5 the the DET fcis-3177 5 6 research research NOUN fcis-3177 5 7 background background NOUN fcis-3177 5 8 and and CCONJ fcis-3177 5 9 definition definition NOUN fcis-3177 5 10 of of ADP fcis-3177 5 11 few few ADJ fcis-3177 5 12 - - PUNCT fcis-3177 5 13 shot shot NOUN fcis-3177 5 14 learning learning NOUN fcis-3177 5 15 , , PUNCT fcis-3177 5 16 introduces introduce VERB fcis-3177 5 17 the the DET fcis-3177 5 18 relevant relevant ADJ fcis-3177 5 19 models model NOUN fcis-3177 5 20 , , PUNCT fcis-3177 5 21 and and CCONJ fcis-3177 5 22 summarizes summarize NOUN fcis-3177 5 23 and and CCONJ fcis-3177 5 24 analyzes analyze VERB fcis-3177 5 25 the the DET fcis-3177 5 26 common common ADJ fcis-3177 5 27 approaches approach NOUN fcis-3177 5 28 to to ADP fcis-3177 5 29 the the DET fcis-3177 5 30 problem problem NOUN fcis-3177 5 31 of of ADP fcis-3177 5 32 few few ADJ fcis-3177 5 33 - - PUNCT fcis-3177 5 34 shot shot NOUN fcis-3177 5 35 learning learning NOUN fcis-3177 5 36 based base VERB fcis-3177 5 37 on on ADP fcis-3177 5 38 deep deep ADJ fcis-3177 5 39 neural neural ADJ fcis-3177 5 40 networks network NOUN fcis-3177 5 41 at at ADP fcis-3177 5 42 the the DET fcis-3177 5 43 present present ADJ fcis-3177 5 44 stage stage NOUN fcis-3177 5 45 , , PUNCT fcis-3177 5 46 which which PRON fcis-3177 5 47 are be AUX fcis-3177 5 48 divided divide VERB fcis-3177 5 49 into into ADP fcis-3177 5 50 four four NUM fcis-3177 5 51 types type NOUN fcis-3177 5 52 : : PUNCT fcis-3177 5 53 data datum NOUN fcis-3177 5 54 augmentation augmentation NOUN fcis-3177 5 55 , , PUNCT fcis-3177 5 56 model model NOUN fcis-3177 5 57 fine fine ADV fcis-3177 5 58 - - PUNCT fcis-3177 5 59 tuning tuning NOUN fcis-3177 5 60 , , PUNCT fcis-3177 5 61 metric metric ADJ fcis-3177 5 62 learning learning NOUN fcis-3177 5 63 and and CCONJ fcis-3177 5 64 meta meta NOUN fcis-3177 5 65 - - PUNCT fcis-3177 5 66 learning learning NOUN fcis-3177 5 67 . . PUNCT fcis-3177 6 1 finally finally ADV fcis-3177 6 2 , , PUNCT fcis-3177 6 3 popular popular ADJ fcis-3177 6 4 datasets dataset NOUN fcis-3177 6 5 for for ADP fcis-3177 6 6 few few ADJ fcis-3177 6 7 - - PUNCT fcis-3177 6 8 shot shot NOUN fcis-3177 6 9 learning learning NOUN fcis-3177 6 10 are be AUX fcis-3177 6 11 described describe VERB fcis-3177 6 12 , , PUNCT fcis-3177 6 13 the the DET fcis-3177 6 14 paper paper NOUN fcis-3177 6 15 is be AUX fcis-3177 6 16 concluded conclude VERB fcis-3177 6 17 and and CCONJ fcis-3177 6 18 future future ADJ fcis-3177 6 19 research research NOUN fcis-3177 6 20 directions direction NOUN fcis-3177 6 21 are be AUX fcis-3177 6 22 discussed discuss VERB fcis-3177 6 23 . . PUNCT fcis-3177 7 1 keywords keyword NOUN fcis-3177 7 2 : : PUNCT fcis-3177 7 3 few few ADJ fcis-3177 7 4 - - PUNCT fcis-3177 7 5 shot shot NOUN fcis-3177 7 6 learning learning NOUN fcis-3177 7 7 ; ; PUNCT fcis-3177 7 8 low low ADJ fcis-3177 7 9 - - PUNCT fcis-3177 7 10 shot shot NOUN fcis-3177 7 11 learning learning NOUN fcis-3177 7 12 ; ; PUNCT fcis-3177 7 13 deep deep ADJ fcis-3177 7 14 learning learning NOUN fcis-3177 7 15 ; ; PUNCT fcis-3177 7 16 deep deep ADJ fcis-3177 7 17 neural neural ADJ fcis-3177 7 18 network network NOUN fcis-3177 7 19 . . PUNCT fcis-3177 8 1 1 1 X fcis-3177 8 2 . . X fcis-3177 8 3 introduction introduction NOUN fcis-3177 8 4 1.1 1.1 NUM fcis-3177 8 5 . . PUNCT fcis-3177 9 1 research research NOUN fcis-3177 9 2 background background NOUN fcis-3177 9 3 and and CCONJ fcis-3177 9 4 significance significance NOUN fcis-3177 9 5 as as ADV fcis-3177 9 6 early early ADV fcis-3177 9 7 as as ADP fcis-3177 9 8 1987 1987 NUM fcis-3177 9 9 , , PUNCT fcis-3177 9 10 research research NOUN fcis-3177 9 11 by by ADP fcis-3177 9 12 biederman biederman NOUN fcis-3177 10 1 [ [ X fcis-3177 10 2 1 1 X fcis-3177 10 3 ] ] PUNCT fcis-3177 10 4 found find VERB fcis-3177 10 5 that that SCONJ fcis-3177 10 6 humans human NOUN fcis-3177 10 7 can can AUX fcis-3177 10 8 identify identify VERB fcis-3177 10 9 an an DET fcis-3177 10 10 average average NOUN fcis-3177 10 11 of of ADP fcis-3177 10 12 10,000 10,000 NUM fcis-3177 10 13 to to PART fcis-3177 10 14 30,000 30,000 NUM fcis-3177 10 15 things thing NOUN fcis-3177 10 16 in in ADP fcis-3177 10 17 our our PRON fcis-3177 10 18 lifetime lifetime NOUN fcis-3177 10 19 . . PUNCT fcis-3177 11 1 first first ADV fcis-3177 11 2 , , PUNCT fcis-3177 11 3 the the DET fcis-3177 11 4 number number NOUN fcis-3177 11 5 of of ADP fcis-3177 11 6 samples sample NOUN fcis-3177 11 7 in in ADP fcis-3177 11 8 the the DET fcis-3177 11 9 real real ADJ fcis-3177 11 10 world world NOUN fcis-3177 11 11 is be AUX fcis-3177 11 12 consistent consistent ADJ fcis-3177 11 13 with with ADP fcis-3177 11 14 a a DET fcis-3177 11 15 long long ADJ fcis-3177 11 16 - - PUNCT fcis-3177 11 17 tail tail NOUN fcis-3177 11 18 distribution distribution NOUN fcis-3177 11 19 [ [ X fcis-3177 11 20 2 2 NUM fcis-3177 11 21 ] ] PUNCT fcis-3177 11 22 , , PUNCT fcis-3177 11 23 as as SCONJ fcis-3177 11 24 shown show VERB fcis-3177 11 25 in in ADP fcis-3177 11 26 figure figure NOUN fcis-3177 11 27 1 1 NUM fcis-3177 11 28 , , PUNCT fcis-3177 11 29 where where SCONJ fcis-3177 11 30 only only ADV fcis-3177 11 31 a a DET fcis-3177 11 32 few few ADJ fcis-3177 11 33 categories category NOUN fcis-3177 11 34 have have VERB fcis-3177 11 35 enough enough ADJ fcis-3177 11 36 samples sample NOUN fcis-3177 11 37 , , PUNCT fcis-3177 11 38 and and CCONJ fcis-3177 11 39 the the DET fcis-3177 11 40 vast vast ADJ fcis-3177 11 41 majority majority NOUN fcis-3177 11 42 have have VERB fcis-3177 11 43 tiny tiny ADJ fcis-3177 11 44 sample sample NOUN fcis-3177 11 45 sizes size NOUN fcis-3177 11 46 . . PUNCT fcis-3177 12 1 moreover moreover ADV fcis-3177 12 2 , , PUNCT fcis-3177 12 3 humans human NOUN fcis-3177 12 4 do do AUX fcis-3177 12 5 not not PART fcis-3177 12 6 need need VERB fcis-3177 12 7 hundreds hundred NOUN fcis-3177 12 8 or or CCONJ fcis-3177 12 9 thousands thousand NOUN fcis-3177 12 10 of of ADP fcis-3177 12 11 data datum NOUN fcis-3177 12 12 when when SCONJ fcis-3177 12 13 learning learn VERB fcis-3177 12 14 a a DET fcis-3177 12 15 new new ADJ fcis-3177 12 16 concept concept NOUN fcis-3177 12 17 . . PUNCT fcis-3177 13 1 this this PRON fcis-3177 13 2 means mean VERB fcis-3177 13 3 that that SCONJ fcis-3177 13 4 people people NOUN fcis-3177 13 5 can can AUX fcis-3177 13 6 learn learn VERB fcis-3177 13 7 quickly quickly ADV fcis-3177 13 8 from from ADP fcis-3177 13 9 a a DET fcis-3177 13 10 small small ADJ fcis-3177 13 11 number number NOUN fcis-3177 13 12 of of ADP fcis-3177 13 13 samples sample NOUN fcis-3177 13 14 and and CCONJ fcis-3177 13 15 use use VERB fcis-3177 13 16 this this PRON fcis-3177 13 17 to to PART fcis-3177 13 18 make make VERB fcis-3177 13 19 summary summary NOUN fcis-3177 13 20 generalizations generalization NOUN fcis-3177 13 21 and and CCONJ fcis-3177 13 22 distinguish distinguish VERB fcis-3177 13 23 between between ADP fcis-3177 13 24 different different ADJ fcis-3177 13 25 samples sample NOUN fcis-3177 13 26 , , PUNCT fcis-3177 13 27 even even ADV fcis-3177 13 28 those those PRON fcis-3177 13 29 they they PRON fcis-3177 13 30 have have AUX fcis-3177 13 31 not not PART fcis-3177 13 32 seen see VERB fcis-3177 13 33 , , PUNCT fcis-3177 13 34 based base VERB fcis-3177 13 35 on on ADP fcis-3177 13 36 prior prior ADJ fcis-3177 13 37 knowledge knowledge NOUN fcis-3177 13 38 . . PUNCT fcis-3177 14 1 figure figure NOUN fcis-3177 14 2 1 1 NUM fcis-3177 14 3 . . PUNCT fcis-3177 15 1 long long ADJ fcis-3177 15 2 - - PUNCT fcis-3177 15 3 tail tail NOUN fcis-3177 15 4 distribution distribution NOUN fcis-3177 15 5 of of ADP fcis-3177 15 6 the the DET fcis-3177 15 7 sample sample NOUN fcis-3177 15 8 deep deep ADJ fcis-3177 15 9 learning learning NOUN fcis-3177 15 10 models model NOUN fcis-3177 15 11 are be AUX fcis-3177 15 12 now now ADV fcis-3177 15 13 widely widely ADV fcis-3177 15 14 used use VERB fcis-3177 15 15 for for ADP fcis-3177 15 16 recognition recognition NOUN fcis-3177 15 17 and and CCONJ fcis-3177 15 18 classification classification NOUN fcis-3177 15 19 tasks task NOUN fcis-3177 15 20 and and CCONJ fcis-3177 15 21 have have AUX fcis-3177 15 22 achieved achieve VERB fcis-3177 15 23 efficient efficient ADJ fcis-3177 15 24 and and CCONJ fcis-3177 15 25 accurate accurate ADJ fcis-3177 15 26 results result NOUN fcis-3177 15 27 , , PUNCT fcis-3177 15 28 but but CCONJ fcis-3177 15 29 for for ADP fcis-3177 15 30 the the DET fcis-3177 15 31 most most ADJ fcis-3177 15 32 part part NOUN fcis-3177 15 33 , , PUNCT fcis-3177 15 34 this this PRON fcis-3177 15 35 is be AUX fcis-3177 15 36 because because SCONJ fcis-3177 15 37 the the DET fcis-3177 15 38 training training NOUN fcis-3177 15 39 of of ADP fcis-3177 15 40 models model NOUN fcis-3177 15 41 often often ADV fcis-3177 15 42 relies rely VERB fcis-3177 15 43 on on ADP fcis-3177 15 44 the the DET fcis-3177 15 45 use use NOUN fcis-3177 15 46 of of ADP fcis-3177 15 47 large large ADJ fcis-3177 15 48 numbers number NOUN fcis-3177 15 49 of of ADP fcis-3177 15 50 labeled label VERB fcis-3177 15 51 datasets dataset NOUN fcis-3177 15 52 ( ( PUNCT fcis-3177 15 53 such such ADJ fcis-3177 15 54 as as ADP fcis-3177 15 55 imagenet imagenet NOUN fcis-3177 15 56 dataset dataset NOUN fcis-3177 15 57 ) ) PUNCT fcis-3177 15 58 . . PUNCT fcis-3177 16 1 however however ADV fcis-3177 16 2 , , PUNCT fcis-3177 16 3 it it PRON fcis-3177 16 4 is be AUX fcis-3177 16 5 often often ADV fcis-3177 16 6 impossible impossible ADJ fcis-3177 16 7 to to PART fcis-3177 16 8 provide provide VERB fcis-3177 16 9 enough enough ADJ fcis-3177 16 10 labeled label VERB fcis-3177 16 11 samples sample NOUN fcis-3177 16 12 in in ADP fcis-3177 16 13 many many ADJ fcis-3177 16 14 real real ADJ fcis-3177 16 15 - - PUNCT fcis-3177 16 16 world world NOUN fcis-3177 16 17 scenarios scenario NOUN fcis-3177 16 18 , , PUNCT fcis-3177 16 19 such such ADJ fcis-3177 16 20 as as ADP fcis-3177 16 21 fault fault NOUN fcis-3177 16 22 diagnosis diagnosis NOUN fcis-3177 16 23 and and CCONJ fcis-3177 16 24 anomaly anomaly NOUN fcis-3177 16 25 detection detection NOUN fcis-3177 16 26 . . PUNCT fcis-3177 17 1 the the DET fcis-3177 17 2 cost cost NOUN fcis-3177 17 3 of of ADP fcis-3177 17 4 labeling labeling NOUN fcis-3177 17 5 samples sample NOUN fcis-3177 17 6 is be AUX fcis-3177 17 7 high high ADJ fcis-3177 17 8 and and CCONJ fcis-3177 17 9 often often ADV fcis-3177 17 10 requires require VERB fcis-3177 17 11 a a DET fcis-3177 17 12 certain certain ADJ fcis-3177 17 13 level level NOUN fcis-3177 17 14 of of ADP fcis-3177 17 15 expertise expertise NOUN fcis-3177 17 16 . . PUNCT fcis-3177 18 1 particularly particularly ADV fcis-3177 18 2 in in ADP fcis-3177 18 3 the the DET fcis-3177 18 4 biological biological ADJ fcis-3177 18 5 and and CCONJ fcis-3177 18 6 medical medical ADJ fcis-3177 18 7 fields field NOUN fcis-3177 18 8 , , PUNCT fcis-3177 18 9 there there PRON fcis-3177 18 10 may may AUX fcis-3177 18 11 be be AUX fcis-3177 18 12 only only ADV fcis-3177 18 13 a a DET fcis-3177 18 14 few few ADJ fcis-3177 18 15 unique unique ADJ fcis-3177 18 16 or or CCONJ fcis-3177 18 17 diseased diseased ADJ fcis-3177 18 18 samples sample NOUN fcis-3177 18 19 for for ADP fcis-3177 18 20 some some DET fcis-3177 18 21 exceptional exceptional ADJ fcis-3177 18 22 cases case NOUN fcis-3177 18 23 or or CCONJ fcis-3177 18 24 rare rare ADJ fcis-3177 18 25 diseases disease NOUN fcis-3177 18 26 , , PUNCT fcis-3177 18 27 which which PRON fcis-3177 18 28 leads lead VERB fcis-3177 18 29 to to ADP fcis-3177 18 30 an an DET fcis-3177 18 31 extreme extreme ADJ fcis-3177 18 32 imbalance imbalance NOUN fcis-3177 18 33 between between ADP fcis-3177 18 34 abnormal abnormal ADJ fcis-3177 18 35 and and CCONJ fcis-3177 18 36 normal normal ADJ fcis-3177 18 37 samples sample NOUN fcis-3177 18 38 . . PUNCT fcis-3177 19 1 insufficient insufficient ADJ fcis-3177 19 2 numbers number NOUN fcis-3177 19 3 of of ADP fcis-3177 19 4 abnormal abnormal ADJ fcis-3177 19 5 samples sample NOUN fcis-3177 19 6 or or CCONJ fcis-3177 19 7 severe severe ADJ fcis-3177 19 8 quality quality NOUN fcis-3177 19 9 bias bias NOUN fcis-3177 19 10 can can AUX fcis-3177 19 11 easily easily ADV fcis-3177 19 12 lead lead VERB fcis-3177 19 13 to to ADP fcis-3177 19 14 a a DET fcis-3177 19 15 situation situation NOUN fcis-3177 19 16 where where SCONJ fcis-3177 19 17 the the PRON fcis-3177 19 18 higher high ADJ fcis-3177 19 19 the the DET fcis-3177 19 20 number number NOUN fcis-3177 19 21 of of ADP fcis-3177 19 22 normal normal ADJ fcis-3177 19 23 samples sample NOUN fcis-3177 19 24 , , PUNCT fcis-3177 19 25 the the PRON fcis-3177 19 26 more more ADV fcis-3177 19 27 the the DET fcis-3177 19 28 neural neural ADJ fcis-3177 19 29 network network NOUN fcis-3177 19 30 's 's PART fcis-3177 19 31 performance performance NOUN fcis-3177 19 32 is be AUX fcis-3177 19 33 affected affect VERB fcis-3177 19 34 . . PUNCT fcis-3177 20 1 1.2 1.2 NUM fcis-3177 20 2 . . PUNCT fcis-3177 20 3 definition definition NOUN fcis-3177 20 4 of of ADP fcis-3177 20 5 few few ADJ fcis-3177 20 6 - - PUNCT fcis-3177 20 7 shot shot NOUN fcis-3177 20 8 learning learning NOUN fcis-3177 20 9 in in ADP fcis-3177 20 10 order order NOUN fcis-3177 20 11 to to PART fcis-3177 20 12 solve solve VERB fcis-3177 20 13 the the DET fcis-3177 20 14 above above ADJ fcis-3177 20 15 problem problem NOUN fcis-3177 20 16 , , PUNCT fcis-3177 20 17 few few ADJ fcis-3177 20 18 - - PUNCT fcis-3177 20 19 shot shot NOUN fcis-3177 20 20 learning learning NOUN fcis-3177 20 21 was be AUX fcis-3177 20 22 born bear VERB fcis-3177 20 23 . . PUNCT fcis-3177 21 1 it it PRON fcis-3177 21 2 was be AUX fcis-3177 21 3 first first ADV fcis-3177 21 4 proposed propose VERB fcis-3177 21 5 by by ADP fcis-3177 21 6 li li PROPN fcis-3177 22 1 [ [ X fcis-3177 22 2 3 3 X fcis-3177 22 3 ] ] PUNCT fcis-3177 22 4 in in ADP fcis-3177 22 5 2003 2003 NUM fcis-3177 22 6 , , PUNCT fcis-3177 22 7 that that PRON fcis-3177 22 8 is be AUX fcis-3177 22 9 learning learn VERB fcis-3177 22 10 a a DET fcis-3177 22 11 new new ADJ fcis-3177 22 12 category category NOUN fcis-3177 22 13 of of ADP fcis-3177 22 14 objects object NOUN fcis-3177 22 15 using use VERB fcis-3177 22 16 only only ADV fcis-3177 22 17 a a DET fcis-3177 22 18 small small ADJ fcis-3177 22 19 number number NOUN fcis-3177 22 20 of of ADP fcis-3177 22 21 training training NOUN fcis-3177 22 22 samples sample NOUN fcis-3177 22 23 of of ADP fcis-3177 22 24 that that DET fcis-3177 22 25 category category NOUN fcis-3177 22 26 . . PUNCT fcis-3177 23 1 meanwhile meanwhile ADV fcis-3177 23 2 , , PUNCT fcis-3177 23 3 we we PRON fcis-3177 23 4 also also ADV fcis-3177 23 5 want want VERB fcis-3177 23 6 the the DET fcis-3177 23 7 machine machine NOUN fcis-3177 23 8 to to PART fcis-3177 23 9 learn learn VERB fcis-3177 23 10 many many ADJ fcis-3177 23 11 base base NOUN fcis-3177 23 12 classes class NOUN fcis-3177 23 13 and and CCONJ fcis-3177 23 14 then then ADV fcis-3177 23 15 learn learn VERB fcis-3177 23 16 a a DET fcis-3177 23 17 new new ADJ fcis-3177 23 18 class class NOUN fcis-3177 23 19 quickly quickly ADV fcis-3177 23 20 with with ADP fcis-3177 23 21 only only ADV fcis-3177 23 22 a a DET fcis-3177 23 23 few few ADJ fcis-3177 23 24 samples sample NOUN fcis-3177 23 25 . . PUNCT fcis-3177 24 1 in in ADP fcis-3177 24 2 general general ADJ fcis-3177 24 3 , , PUNCT fcis-3177 24 4 few few ADJ fcis-3177 24 5 - - PUNCT fcis-3177 24 6 shot shot NOUN fcis-3177 24 7 learning learning NOUN fcis-3177 24 8 can can AUX fcis-3177 24 9 be be AUX fcis-3177 24 10 done do VERB fcis-3177 24 11 using use VERB fcis-3177 24 12 a a DET fcis-3177 24 13 small small ADJ fcis-3177 24 14 number number NOUN fcis-3177 24 15 of of ADP fcis-3177 24 16 samples sample NOUN fcis-3177 24 17 ( ( PUNCT fcis-3177 24 18 one one NUM fcis-3177 24 19 or or CCONJ fcis-3177 24 20 a a DET fcis-3177 24 21 few few ADJ fcis-3177 24 22 ) ) PUNCT fcis-3177 24 23 from from ADP fcis-3177 24 24 a a DET fcis-3177 24 25 category category NOUN fcis-3177 24 26 . . PUNCT fcis-3177 25 1 in in ADP fcis-3177 25 2 the the DET fcis-3177 25 3 case case NOUN fcis-3177 25 4 of of ADP fcis-3177 25 5 a a DET fcis-3177 25 6 single single ADJ fcis-3177 25 7 training training NOUN fcis-3177 25 8 sample sample NOUN fcis-3177 25 9 , , PUNCT fcis-3177 25 10 it it PRON fcis-3177 25 11 is be AUX fcis-3177 25 12 also also ADV fcis-3177 25 13 known know VERB fcis-3177 25 14 as as ADP fcis-3177 25 15 one one NUM fcis-3177 25 16 - - PUNCT fcis-3177 25 17 shot shot NOUN fcis-3177 25 18 learning learning NOUN fcis-3177 25 19 . . PUNCT fcis-3177 26 1 take take VERB fcis-3177 26 2 the the DET fcis-3177 26 3 classification classification NOUN fcis-3177 26 4 problem problem NOUN fcis-3177 26 5 as as ADP fcis-3177 26 6 an an DET fcis-3177 26 7 example example NOUN fcis-3177 26 8 , , PUNCT fcis-3177 26 9 as as SCONJ fcis-3177 26 10 shown show VERB fcis-3177 26 11 in in ADP fcis-3177 26 12 figure figure NOUN fcis-3177 26 13 2 2 NUM fcis-3177 26 14 below below ADV fcis-3177 26 15 . . PUNCT fcis-3177 27 1 the the DET fcis-3177 27 2 basic basic ADJ fcis-3177 27 3 model model NOUN fcis-3177 27 4 of of ADP fcis-3177 27 5 few few ADJ fcis-3177 27 6 - - PUNCT fcis-3177 27 7 shot shot NOUN fcis-3177 27 8 learning learning NOUN fcis-3177 27 9 can can AUX fcis-3177 27 10 be be AUX fcis-3177 27 11 defined define VERB fcis-3177 27 12 as as ADP fcis-3177 27 13 𝑝 𝑝 NOUN fcis-3177 27 14 = = SYM fcis-3177 27 15 𝐢(𝑓(π‘₯|πœƒ)|𝑀 𝐢(𝑓(π‘₯|πœƒ)|𝑀 NOUN fcis-3177 27 16 ) ) PUNCT fcis-3177 27 17 , , PUNCT fcis-3177 27 18 which which PRON fcis-3177 27 19 consists consist VERB fcis-3177 27 20 of of ADP fcis-3177 27 21 a a DET fcis-3177 27 22 feature feature NOUN fcis-3177 27 23 extractor extractor NOUN fcis-3177 27 24 𝑓(βˆ™ 𝑓(βˆ™ NOUN fcis-3177 27 25 |πœƒ |πœƒ NOUN fcis-3177 27 26 ) ) PUNCT fcis-3177 27 27 and and CCONJ fcis-3177 27 28 a a DET fcis-3177 27 29 classifier classifier NOUN fcis-3177 27 30 𝐢(βˆ™ 𝐢(βˆ™ VERB fcis-3177 27 31 |𝑀 |𝑀 NOUN fcis-3177 27 32 ) ) PUNCT fcis-3177 27 33 . . PUNCT fcis-3177 28 1 𝑓(π‘₯|πœƒ 𝑓(π‘₯|πœƒ PROPN fcis-3177 28 2 ) ) PUNCT fcis-3177 28 3 indicates indicate VERB fcis-3177 28 4 the the DET fcis-3177 28 5 features feature NOUN fcis-3177 28 6 extracted extract VERB fcis-3177 28 7 for for ADP fcis-3177 28 8 x x PROPN fcis-3177 28 9 , , PUNCT fcis-3177 28 10 ΞΈ ΞΈ PROPN fcis-3177 28 11 and and CCONJ fcis-3177 28 12 w w PROPN fcis-3177 28 13 indicate indicate VERB fcis-3177 28 14 the the DET fcis-3177 28 15 corresponding correspond VERB fcis-3177 28 16 parameters parameter NOUN fcis-3177 28 17 , , PUNCT fcis-3177 28 18 and and CCONJ fcis-3177 28 19 finally finally ADV fcis-3177 28 20 the the DET fcis-3177 28 21 classification classification NOUN fcis-3177 28 22 prediction prediction NOUN fcis-3177 28 23 result result NOUN fcis-3177 28 24 of of ADP fcis-3177 28 25 the the DET fcis-3177 28 26 sample sample NOUN fcis-3177 28 27 is be AUX fcis-3177 28 28 obtained obtain VERB fcis-3177 28 29 . . PUNCT fcis-3177 29 1 figure figure NOUN fcis-3177 29 2 2 2 NUM fcis-3177 29 3 . . PUNCT fcis-3177 29 4 basic basic ADJ fcis-3177 29 5 model model NOUN fcis-3177 29 6 of of ADP fcis-3177 29 7 few few ADJ fcis-3177 29 8 - - PUNCT fcis-3177 29 9 shot shot NOUN fcis-3177 29 10 learning learning NOUN fcis-3177 29 11 assume assume VERB fcis-3177 29 12 that that SCONJ fcis-3177 29 13 there there PRON fcis-3177 29 14 are be VERB fcis-3177 29 15 n n DET fcis-3177 29 16 classifications classification NOUN fcis-3177 29 17 in in ADP fcis-3177 29 18 our our PRON fcis-3177 29 19 support support NOUN fcis-3177 29 20 set set VERB fcis-3177 29 21 , , PUNCT fcis-3177 29 22 with with ADP fcis-3177 29 23 k k PROPN fcis-3177 29 24 samples sample NOUN fcis-3177 29 25 in in ADP fcis-3177 29 26 each each DET fcis-3177 29 27 classification classification NOUN fcis-3177 29 28 , , PUNCT fcis-3177 29 29 so so SCONJ fcis-3177 29 30 that that SCONJ fcis-3177 29 31 there there PRON fcis-3177 29 32 is be VERB fcis-3177 29 33 a a DET fcis-3177 29 34 total total NOUN fcis-3177 29 35 of of ADP fcis-3177 29 36 n n NUM fcis-3177 29 37 Γ— Γ— PROPN fcis-3177 29 38 k k PROPN fcis-3177 29 39 samples sample NOUN fcis-3177 29 40 . . PUNCT fcis-3177 30 1 the the DET fcis-3177 30 2 task task NOUN fcis-3177 30 3 of of ADP fcis-3177 30 4 training training NOUN fcis-3177 30 5 a a DET fcis-3177 30 6 model model NOUN fcis-3177 30 7 that that PRON fcis-3177 30 8 can can AUX fcis-3177 30 9 distinguish distinguish VERB fcis-3177 30 10 between between ADP fcis-3177 30 11 n n NOUN fcis-3177 30 12 classifications classification NOUN fcis-3177 30 13 from from ADP fcis-3177 30 14 n n PRON fcis-3177 30 15 Γ— Γ— NOUN fcis-3177 30 16 k k PROPN fcis-3177 30 17 samples sample NOUN fcis-3177 30 18 is be AUX fcis-3177 30 19 defined define VERB fcis-3177 30 20 as as ADP fcis-3177 30 21 the the DET fcis-3177 30 22 n n NUM fcis-3177 30 23 - - PUNCT fcis-3177 30 24 way way NOUN fcis-3177 30 25 k k NOUN fcis-3177 30 26 - - PUNCT fcis-3177 30 27 shot shoot VERB fcis-3177 30 28 problem problem NOUN fcis-3177 30 29 [ [ X fcis-3177 30 30 4 4 NUM fcis-3177 30 31 ] ] PUNCT fcis-3177 30 32 . . PUNCT fcis-3177 31 1 because because SCONJ fcis-3177 31 2 it it PRON fcis-3177 31 3 is be AUX fcis-3177 31 4 a a DET fcis-3177 31 5 fewshot fewshot ADJ fcis-3177 31 6 learning learning NOUN fcis-3177 31 7 , , PUNCT fcis-3177 31 8 the the DET fcis-3177 31 9 value value NOUN fcis-3177 31 10 of of ADP fcis-3177 31 11 k k PROPN fcis-3177 31 12 is be AUX fcis-3177 31 13 generally generally ADV fcis-3177 31 14 small small ADJ fcis-3177 31 15 ( ( PUNCT fcis-3177 31 16 usually usually ADV fcis-3177 31 17 between between ADP fcis-3177 31 18 1 1 NUM fcis-3177 31 19 and and CCONJ fcis-3177 31 20 20 20 NUM fcis-3177 31 21 ) ) PUNCT fcis-3177 31 22 . . PUNCT fcis-3177 32 1 if if SCONJ fcis-3177 32 2 there there PRON fcis-3177 32 3 are be VERB fcis-3177 32 4 n n PRON fcis-3177 32 5 classifications classification NOUN fcis-3177 32 6 in in ADP fcis-3177 32 7 the the DET fcis-3177 32 8 support support NOUN fcis-3177 32 9 set set VERB fcis-3177 32 10 s s PROPN fcis-3177 32 11 , , PUNCT fcis-3177 32 12 where where SCONJ fcis-3177 32 13 there there PRON fcis-3177 32 14 are be VERB fcis-3177 32 15 k k PROPN fcis-3177 32 16 samples sample NOUN fcis-3177 32 17 xi xi X fcis-3177 32 18 in in ADP fcis-3177 32 19 each each DET fcis-3177 32 20 classification classification NOUN fcis-3177 32 21 , , PUNCT fcis-3177 32 22 each each PRON fcis-3177 32 23 with with ADP fcis-3177 32 24 a a DET fcis-3177 32 25 different different ADJ fcis-3177 32 26 label label NOUN fcis-3177 32 27 yi yi NOUN fcis-3177 32 28 , , PUNCT fcis-3177 32 29 then then ADV fcis-3177 32 30 this this PRON fcis-3177 32 31 can can AUX fcis-3177 32 32 be be AUX fcis-3177 32 33 expressed express VERB fcis-3177 32 34 as as SCONJ fcis-3177 32 35 equation equation NOUN fcis-3177 32 36 ( ( PUNCT fcis-3177 32 37 1 1 NUM fcis-3177 32 38 ): ): PUNCT fcis-3177 32 39 𝑆 𝑆 PROPN fcis-3177 32 40 = = SYM fcis-3177 32 41 { { PUNCT fcis-3177 32 42 ( ( PUNCT fcis-3177 32 43 π‘₯1 π‘₯1 NOUN fcis-3177 32 44 , , PUNCT fcis-3177 32 45 𝑦1),β‹― 𝑦1),β‹― PROPN fcis-3177 32 46 , , PUNCT fcis-3177 32 47 ( ( PUNCT fcis-3177 32 48 π‘₯𝐾×𝑁 π‘₯𝐾×𝑁 X fcis-3177 32 49 , , PUNCT fcis-3177 32 50 𝑦𝐾×𝑁 𝑦𝐾×𝑁 NOUN fcis-3177 32 51 ) ) PUNCT fcis-3177 32 52 } } PUNCT fcis-3177 32 53 111 111 NUM fcis-3177 32 54 similarly similarly ADV fcis-3177 32 55 , , PUNCT fcis-3177 32 56 based base VERB fcis-3177 32 57 on on ADP fcis-3177 32 58 the the DET fcis-3177 32 59 category category NOUN fcis-3177 32 60 labels label VERB fcis-3177 32 61 in in ADP fcis-3177 32 62 the the DET fcis-3177 32 63 support support NOUN fcis-3177 32 64 set set NOUN fcis-3177 32 65 , , PUNCT fcis-3177 32 66 the the DET fcis-3177 32 67 query query NOUN fcis-3177 32 68 set set VERB fcis-3177 32 69 q q NOUN fcis-3177 32 70 consisting consisting NOUN fcis-3177 32 71 of of ADP fcis-3177 32 72 m m PRON fcis-3177 32 73 randomly randomly ADV fcis-3177 32 74 sampled sample VERB fcis-3177 32 75 samples sample NOUN fcis-3177 32 76 that that PRON fcis-3177 32 77 do do AUX fcis-3177 32 78 not not PART fcis-3177 32 79 duplicate duplicate VERB fcis-3177 32 80 the the DET fcis-3177 32 81 support support NOUN fcis-3177 32 82 set set NOUN fcis-3177 32 83 is be AUX fcis-3177 32 84 expressed express VERB fcis-3177 32 85 as as ADP fcis-3177 32 86 equation equation NOUN fcis-3177 32 87 ( ( PUNCT fcis-3177 32 88 2 2 NUM fcis-3177 32 89 ): ): PUNCT fcis-3177 33 1 𝑄 𝑄 PROPN fcis-3177 33 2 = = SYM fcis-3177 33 3 { { PUNCT fcis-3177 33 4 ( ( PUNCT fcis-3177 33 5 π‘₯1 π‘₯1 NOUN fcis-3177 33 6 , , PUNCT fcis-3177 33 7 𝑦1),β‹― 𝑦1),β‹― PROPN fcis-3177 33 8 , , PUNCT fcis-3177 33 9 ( ( PUNCT fcis-3177 33 10 π‘₯𝑀 π‘₯𝑀 NOUN fcis-3177 33 11 , , PUNCT fcis-3177 33 12 𝑦𝑀 𝑦𝑀 NOUN fcis-3177 33 13 ) ) PUNCT fcis-3177 33 14 } } PUNCT fcis-3177 33 15 2 2 NUM fcis-3177 33 16 . . X fcis-3177 33 17 deep deep ADJ fcis-3177 33 18 learning learn VERB fcis-3177 33 19 related relate VERB fcis-3177 33 20 models model NOUN fcis-3177 33 21 in in ADP fcis-3177 33 22 2006 2006 NUM fcis-3177 33 23 , , PUNCT fcis-3177 33 24 hinton hinton PROPN fcis-3177 33 25 et et PROPN fcis-3177 33 26 al al PROPN fcis-3177 33 27 . . PUNCT fcis-3177 34 1 [ [ X fcis-3177 34 2 5 5 NUM fcis-3177 34 3 ] ] PUNCT fcis-3177 34 4 proposed propose VERB fcis-3177 34 5 that that SCONJ fcis-3177 34 6 an an DET fcis-3177 34 7 artificial artificial ADJ fcis-3177 34 8 neural neural ADJ fcis-3177 34 9 network network NOUN fcis-3177 34 10 with with ADP fcis-3177 34 11 multiple multiple ADJ fcis-3177 34 12 hidden hide VERB fcis-3177 34 13 layers layer NOUN fcis-3177 34 14 has have VERB fcis-3177 34 15 excellent excellent ADJ fcis-3177 34 16 feature feature NOUN fcis-3177 34 17 learning learn VERB fcis-3177 34 18 ability ability NOUN fcis-3177 34 19 and and CCONJ fcis-3177 34 20 that that DET fcis-3177 34 21 layerwise layerwise VERB fcis-3177 34 22 pre pre ADJ fcis-3177 34 23 - - NOUN fcis-3177 34 24 training training NOUN fcis-3177 34 25 can can AUX fcis-3177 34 26 effectively effectively ADV fcis-3177 34 27 overcome overcome VERB fcis-3177 34 28 the the DET fcis-3177 34 29 difficulties difficulty NOUN fcis-3177 34 30 in in ADP fcis-3177 34 31 training train VERB fcis-3177 34 32 the the DET fcis-3177 34 33 deep deep ADJ fcis-3177 34 34 neural neural ADJ fcis-3177 34 35 network network NOUN fcis-3177 34 36 , , PUNCT fcis-3177 34 37 which which PRON fcis-3177 34 38 has have AUX fcis-3177 34 39 led lead VERB fcis-3177 34 40 to to ADP fcis-3177 34 41 the the DET fcis-3177 34 42 research research NOUN fcis-3177 34 43 of of ADP fcis-3177 34 44 deep deep ADJ fcis-3177 34 45 learning learning NOUN fcis-3177 34 46 . . PUNCT fcis-3177 35 1 with with ADP fcis-3177 35 2 continuous continuous ADJ fcis-3177 35 3 research research NOUN fcis-3177 35 4 on on ADP fcis-3177 35 5 deep deep ADJ fcis-3177 35 6 learning learning NOUN fcis-3177 35 7 theory theory NOUN fcis-3177 35 8 and and CCONJ fcis-3177 35 9 upgrading upgrade VERB fcis-3177 35 10 numerical numerical ADJ fcis-3177 35 11 computing computing PROPN fcis-3177 35 12 equipment equipment NOUN fcis-3177 35 13 , , PUNCT fcis-3177 35 14 dozens dozen NOUN fcis-3177 35 15 of of ADP fcis-3177 35 16 deep deep ADJ fcis-3177 35 17 learning learning NOUN fcis-3177 35 18 models model NOUN fcis-3177 35 19 have have AUX fcis-3177 35 20 been be AUX fcis-3177 35 21 developed develop VERB fcis-3177 35 22 and and CCONJ fcis-3177 35 23 applied apply VERB fcis-3177 35 24 to to ADP fcis-3177 35 25 different different ADJ fcis-3177 35 26 fields field NOUN fcis-3177 35 27 . . PUNCT fcis-3177 36 1 few few ADJ fcis-3177 36 2 - - PUNCT fcis-3177 36 3 shot shot NOUN fcis-3177 36 4 learning learning NOUN fcis-3177 36 5 based base VERB fcis-3177 36 6 on on ADP fcis-3177 36 7 deep deep ADJ fcis-3177 36 8 learning learning NOUN fcis-3177 36 9 models model NOUN fcis-3177 36 10 is be AUX fcis-3177 36 11 also also ADV fcis-3177 36 12 a a DET fcis-3177 36 13 hot hot ADJ fcis-3177 36 14 research research NOUN fcis-3177 36 15 topic topic NOUN fcis-3177 36 16 in in ADP fcis-3177 36 17 recent recent ADJ fcis-3177 36 18 years year NOUN fcis-3177 36 19 . . PUNCT fcis-3177 37 1 the the DET fcis-3177 37 2 most most ADV fcis-3177 37 3 typical typical ADJ fcis-3177 37 4 deep deep ADJ fcis-3177 37 5 neural neural ADJ fcis-3177 37 6 networks network NOUN fcis-3177 37 7 used use VERB fcis-3177 37 8 for for ADP fcis-3177 37 9 few few ADJ fcis-3177 37 10 - - PUNCT fcis-3177 37 11 shot shot NOUN fcis-3177 37 12 learning learning NOUN fcis-3177 37 13 are be AUX fcis-3177 37 14 the the DET fcis-3177 37 15 convolutional convolutional ADJ fcis-3177 37 16 neural neural ADJ fcis-3177 37 17 network network NOUN fcis-3177 37 18 ( ( PUNCT fcis-3177 37 19 cnn cnn PROPN fcis-3177 37 20 ) ) PUNCT fcis-3177 37 21 and and CCONJ fcis-3177 37 22 recurrent recurrent ADJ fcis-3177 37 23 neural neural ADJ fcis-3177 37 24 network network NOUN fcis-3177 37 25 ( ( PUNCT fcis-3177 37 26 rnn rnn PROPN fcis-3177 37 27 ) ) PUNCT fcis-3177 37 28 , , PUNCT fcis-3177 37 29 which which PRON fcis-3177 37 30 are be AUX fcis-3177 37 31 the the DET fcis-3177 37 32 most most ADV fcis-3177 37 33 widely widely ADV fcis-3177 37 34 used use VERB fcis-3177 37 35 deep deep ADJ fcis-3177 37 36 learning learning NOUN fcis-3177 37 37 models model NOUN fcis-3177 37 38 . . PUNCT fcis-3177 38 1 most most ADJ fcis-3177 38 2 deep deep ADJ fcis-3177 38 3 learning learning NOUN fcis-3177 38 4 models model NOUN fcis-3177 38 5 used use VERB fcis-3177 38 6 in in ADP fcis-3177 38 7 few few ADJ fcis-3177 38 8 - - PUNCT fcis-3177 38 9 shot shot NOUN fcis-3177 38 10 learning learning NOUN fcis-3177 38 11 are be AUX fcis-3177 38 12 variants variant NOUN fcis-3177 38 13 of of ADP fcis-3177 38 14 these these DET fcis-3177 38 15 two two NUM fcis-3177 38 16 models model NOUN fcis-3177 38 17 , , PUNCT fcis-3177 38 18 such such ADJ fcis-3177 38 19 as as ADP fcis-3177 38 20 resnets resnet NOUN fcis-3177 38 21 [ [ X fcis-3177 38 22 6 6 NUM fcis-3177 38 23 ] ] PUNCT fcis-3177 38 24 , , PUNCT fcis-3177 38 25 googlenet googlenet NOUN fcis-3177 38 26 [ [ X fcis-3177 38 27 7 7 NUM fcis-3177 38 28 ] ] PUNCT fcis-3177 38 29 , , PUNCT fcis-3177 38 30 and and CCONJ fcis-3177 38 31 vgg vgg NOUN fcis-3177 39 1 [ [ X fcis-3177 39 2 8 8 NUM fcis-3177 39 3 ] ] PUNCT fcis-3177 39 4 . . PUNCT fcis-3177 40 1 2.1 2.1 NUM fcis-3177 40 2 . . PUNCT fcis-3177 41 1 convolutional convolutional ADJ fcis-3177 41 2 neural neural ADJ fcis-3177 41 3 network network NOUN fcis-3177 41 4 the the DET fcis-3177 41 5 basic basic ADJ fcis-3177 41 6 structure structure NOUN fcis-3177 41 7 of of ADP fcis-3177 41 8 a a DET fcis-3177 41 9 convolutional convolutional ADJ fcis-3177 41 10 neural neural ADJ fcis-3177 41 11 network network NOUN fcis-3177 41 12 consists consist VERB fcis-3177 41 13 of of ADP fcis-3177 41 14 the the DET fcis-3177 41 15 input input NOUN fcis-3177 41 16 layer layer NOUN fcis-3177 41 17 , , PUNCT fcis-3177 41 18 the the DET fcis-3177 41 19 convolutional convolutional ADJ fcis-3177 41 20 layer layer NOUN fcis-3177 41 21 , , PUNCT fcis-3177 41 22 the the DET fcis-3177 41 23 pooling pool VERB fcis-3177 41 24 layer layer NOUN fcis-3177 41 25 , , PUNCT fcis-3177 41 26 the the DET fcis-3177 41 27 fully fully ADV fcis-3177 41 28 - - PUNCT fcis-3177 41 29 connected connect VERB fcis-3177 41 30 layer layer NOUN fcis-3177 41 31 , , PUNCT fcis-3177 41 32 and and CCONJ fcis-3177 41 33 the the DET fcis-3177 41 34 output output NOUN fcis-3177 41 35 layer layer NOUN fcis-3177 41 36 . . PUNCT fcis-3177 42 1 the the DET fcis-3177 42 2 convolutional convolutional ADJ fcis-3177 42 3 and and CCONJ fcis-3177 42 4 pooling pool VERB fcis-3177 42 5 layers layer NOUN fcis-3177 42 6 are be AUX fcis-3177 42 7 generally generally ADV fcis-3177 42 8 taken take VERB fcis-3177 42 9 in in ADP fcis-3177 42 10 multiple multiple NOUN fcis-3177 42 11 . . PUNCT fcis-3177 43 1 they they PRON fcis-3177 43 2 are be AUX fcis-3177 43 3 connected connect VERB fcis-3177 43 4 in in ADP fcis-3177 43 5 an an DET fcis-3177 43 6 alternating alternate VERB fcis-3177 43 7 design design NOUN fcis-3177 43 8 , , PUNCT fcis-3177 43 9 which which PRON fcis-3177 43 10 forms form VERB fcis-3177 43 11 the the DET fcis-3177 43 12 core core NOUN fcis-3177 43 13 module module NOUN fcis-3177 43 14 of of ADP fcis-3177 43 15 the the DET fcis-3177 43 16 convolutional convolutional ADJ fcis-3177 43 17 neural neural ADJ fcis-3177 43 18 network network NOUN fcis-3177 43 19 , , PUNCT fcis-3177 43 20 while while SCONJ fcis-3177 43 21 the the DET fcis-3177 43 22 higher high ADJ fcis-3177 43 23 layers layer NOUN fcis-3177 43 24 are be AUX fcis-3177 43 25 made make VERB fcis-3177 43 26 up up ADP fcis-3177 43 27 of of ADP fcis-3177 43 28 fully fully ADV fcis-3177 43 29 connected connected ADJ fcis-3177 43 30 layers layer NOUN fcis-3177 43 31 . . PUNCT fcis-3177 44 1 figure figure VERB fcis-3177 44 2 3 3 NUM fcis-3177 44 3 . . PUNCT fcis-3177 44 4 basic basic ADJ fcis-3177 44 5 structure structure NOUN fcis-3177 44 6 of of ADP fcis-3177 44 7 convolutional convolutional ADJ fcis-3177 44 8 neural neural ADJ fcis-3177 44 9 network network NOUN fcis-3177 44 10 in in ADP fcis-3177 44 11 the the DET fcis-3177 44 12 convolutional convolutional ADJ fcis-3177 44 13 neural neural ADJ fcis-3177 44 14 network network NOUN fcis-3177 44 15 , , PUNCT fcis-3177 44 16 the the DET fcis-3177 44 17 most most ADV fcis-3177 44 18 important important ADJ fcis-3177 44 19 is be AUX fcis-3177 44 20 the the DET fcis-3177 44 21 convolutional convolutional ADJ fcis-3177 44 22 layer layer NOUN fcis-3177 44 23 , , PUNCT fcis-3177 44 24 which which PRON fcis-3177 44 25 is be AUX fcis-3177 44 26 responsible responsible ADJ fcis-3177 44 27 for for ADP fcis-3177 44 28 extracting extract VERB fcis-3177 44 29 feature feature NOUN fcis-3177 44 30 information information NOUN fcis-3177 44 31 from from ADP fcis-3177 44 32 the the DET fcis-3177 44 33 input input NOUN fcis-3177 44 34 data datum NOUN fcis-3177 44 35 . . PUNCT fcis-3177 45 1 it it PRON fcis-3177 45 2 consists consist VERB fcis-3177 45 3 of of ADP fcis-3177 45 4 several several ADJ fcis-3177 45 5 feature feature NOUN fcis-3177 45 6 maps map NOUN fcis-3177 45 7 , , PUNCT fcis-3177 45 8 each each PRON fcis-3177 45 9 of of ADP fcis-3177 45 10 which which PRON fcis-3177 45 11 consists consist VERB fcis-3177 45 12 of of ADP fcis-3177 45 13 several several ADJ fcis-3177 45 14 neurons neuron NOUN fcis-3177 45 15 . . PUNCT fcis-3177 46 1 each each DET fcis-3177 46 2 neuron neuron NOUN fcis-3177 46 3 of of ADP fcis-3177 46 4 the the DET fcis-3177 46 5 output output NOUN fcis-3177 46 6 feature feature NOUN fcis-3177 46 7 map map NOUN fcis-3177 46 8 in in ADP fcis-3177 46 9 the the DET fcis-3177 46 10 convolutional convolutional ADJ fcis-3177 46 11 layer layer NOUN fcis-3177 46 12 is be AUX fcis-3177 46 13 locally locally ADV fcis-3177 46 14 connected connect VERB fcis-3177 46 15 to to ADP fcis-3177 46 16 its its PRON fcis-3177 46 17 input input NOUN fcis-3177 46 18 . . PUNCT fcis-3177 47 1 the the DET fcis-3177 47 2 corresponding corresponding ADJ fcis-3177 47 3 connection connection NOUN fcis-3177 47 4 weights weight NOUN fcis-3177 47 5 are be AUX fcis-3177 47 6 weighted weight VERB fcis-3177 47 7 and and CCONJ fcis-3177 47 8 summed sum VERB fcis-3177 47 9 with with ADP fcis-3177 47 10 the the DET fcis-3177 47 11 local local ADJ fcis-3177 47 12 input input NOUN fcis-3177 47 13 plus plus CCONJ fcis-3177 47 14 a a DET fcis-3177 47 15 bias bias NOUN fcis-3177 47 16 value value NOUN fcis-3177 47 17 to to PART fcis-3177 47 18 obtain obtain VERB fcis-3177 47 19 the the DET fcis-3177 47 20 input input NOUN fcis-3177 47 21 value value NOUN fcis-3177 47 22 of of ADP fcis-3177 47 23 that that DET fcis-3177 47 24 neuron neuron NOUN fcis-3177 47 25 , , PUNCT fcis-3177 47 26 the the DET fcis-3177 47 27 process process NOUN fcis-3177 47 28 is be AUX fcis-3177 47 29 equivalent equivalent ADJ fcis-3177 47 30 to to ADP fcis-3177 47 31 the the DET fcis-3177 47 32 convolution convolution NOUN fcis-3177 47 33 process process NOUN fcis-3177 47 34 , , PUNCT fcis-3177 47 35 from from ADP fcis-3177 47 36 which which PRON fcis-3177 47 37 the the DET fcis-3177 47 38 cnn cnn PROPN fcis-3177 47 39 gets get VERB fcis-3177 47 40 its its PRON fcis-3177 47 41 name name NOUN fcis-3177 48 1 [ [ X fcis-3177 48 2 9 9 NUM fcis-3177 48 3 ] ] PUNCT fcis-3177 48 4 . . PUNCT fcis-3177 49 1 in in ADP fcis-3177 49 2 equation equation NOUN fcis-3177 49 3 ( ( PUNCT fcis-3177 49 4 3 3 NUM fcis-3177 49 5 ) ) PUNCT fcis-3177 49 6 , , PUNCT fcis-3177 49 7 f(x f(x PROPN fcis-3177 49 8 ) ) PUNCT fcis-3177 49 9 represents represent VERB fcis-3177 49 10 the the DET fcis-3177 49 11 output output NOUN fcis-3177 49 12 feature feature NOUN fcis-3177 49 13 , , PUNCT fcis-3177 49 14 ΞΈi ΞΈi PROPN fcis-3177 49 15 , , PUNCT fcis-3177 49 16 j j PROPN fcis-3177 49 17 represents represent VERB fcis-3177 49 18 the the DET fcis-3177 49 19 size size NOUN fcis-3177 49 20 of of ADP fcis-3177 49 21 the the DET fcis-3177 49 22 convolution convolution NOUN fcis-3177 49 23 kernel kernel NOUN fcis-3177 49 24 elements element NOUN fcis-3177 49 25 in in ADP fcis-3177 49 26 row row NOUN fcis-3177 50 1 i i PRON fcis-3177 50 2 and and CCONJ fcis-3177 50 3 column column PROPN fcis-3177 50 4 j j PROPN fcis-3177 50 5 , , PUNCT fcis-3177 50 6 xi xi PROPN fcis-3177 50 7 , , PUNCT fcis-3177 50 8 j j PROPN fcis-3177 50 9 represents represent VERB fcis-3177 50 10 the the DET fcis-3177 50 11 size size NOUN fcis-3177 50 12 of of ADP fcis-3177 50 13 the the DET fcis-3177 50 14 elements element NOUN fcis-3177 50 15 in in ADP fcis-3177 50 16 row row NOUN fcis-3177 51 1 i i PRON fcis-3177 51 2 and and CCONJ fcis-3177 51 3 column column PROPN fcis-3177 51 4 j j PROPN fcis-3177 51 5 , , PUNCT fcis-3177 51 6 and and CCONJ fcis-3177 51 7 b b NOUN fcis-3177 51 8 is be AUX fcis-3177 51 9 the the DET fcis-3177 51 10 bias bias NOUN fcis-3177 51 11 . . PUNCT fcis-3177 52 1 the the DET fcis-3177 52 2 convolutional convolutional ADJ fcis-3177 52 3 layer layer NOUN fcis-3177 52 4 has have VERB fcis-3177 52 5 the the DET fcis-3177 52 6 characteristics characteristic NOUN fcis-3177 52 7 of of ADP fcis-3177 52 8 local local ADJ fcis-3177 52 9 receptive receptive ADJ fcis-3177 52 10 field field NOUN fcis-3177 52 11 and and CCONJ fcis-3177 52 12 weight weight NOUN fcis-3177 52 13 sharing sharing NOUN fcis-3177 52 14 , , PUNCT fcis-3177 52 15 which which PRON fcis-3177 52 16 can can AUX fcis-3177 52 17 reduce reduce VERB fcis-3177 52 18 the the DET fcis-3177 52 19 parameters parameter NOUN fcis-3177 52 20 in in ADP fcis-3177 52 21 the the DET fcis-3177 52 22 network network NOUN fcis-3177 52 23 . . PUNCT fcis-3177 53 1 𝑓(π‘₯ 𝑓(π‘₯ NOUN fcis-3177 53 2 ) ) PUNCT fcis-3177 54 1 = = NOUN fcis-3177 54 2 βˆ‘πœƒπ‘–,𝑗 βˆ‘πœƒπ‘–,𝑗 NOUN fcis-3177 54 3 𝑛 𝑛 VERB fcis-3177 54 4 𝑖,𝑗 𝑖,𝑗 ADJ fcis-3177 54 5 Γ— Γ— NOUN fcis-3177 54 6 π‘₯𝑖,𝑗 π‘₯𝑖,𝑗 NOUN fcis-3177 54 7 + + CCONJ fcis-3177 54 8 𝑏 𝑏 DET fcis-3177 54 9 2.2 2.2 NUM fcis-3177 54 10 . . PUNCT fcis-3177 55 1 recurrent recurrent ADJ fcis-3177 55 2 neural neural ADJ fcis-3177 55 3 network network NOUN fcis-3177 55 4 recurrent recurrent NOUN fcis-3177 55 5 neural neural ADJ fcis-3177 55 6 networks network NOUN fcis-3177 55 7 are be AUX fcis-3177 55 8 a a DET fcis-3177 55 9 special special ADJ fcis-3177 55 10 type type NOUN fcis-3177 55 11 of of ADP fcis-3177 55 12 neural neural ADJ fcis-3177 55 13 network network NOUN fcis-3177 55 14 in in ADP fcis-3177 55 15 deep deep ADJ fcis-3177 55 16 learning learning NOUN fcis-3177 55 17 that that PRON fcis-3177 55 18 are be AUX fcis-3177 55 19 internally internally ADV fcis-3177 55 20 self self NOUN fcis-3177 55 21 - - PUNCT fcis-3177 55 22 connected connect VERB fcis-3177 55 23 and and CCONJ fcis-3177 55 24 can can AUX fcis-3177 55 25 learn learn VERB fcis-3177 55 26 complex complex ADJ fcis-3177 55 27 vector vector NOUN fcis-3177 55 28 - - PUNCT fcis-3177 55 29 to to ADP fcis-3177 55 30 - - PUNCT fcis-3177 55 31 vector vector NOUN fcis-3177 55 32 mappings mapping NOUN fcis-3177 55 33 . . PUNCT fcis-3177 56 1 the the DET fcis-3177 56 2 first first ADJ fcis-3177 56 3 research research NOUN fcis-3177 56 4 on on ADP fcis-3177 56 5 rnns rnns PROPN fcis-3177 56 6 was be AUX fcis-3177 56 7 proposed propose VERB fcis-3177 56 8 by by ADP fcis-3177 56 9 hopfield hopfield PROPN fcis-3177 56 10 with with ADP fcis-3177 56 11 the the DET fcis-3177 56 12 hopfield hopfield PROPN fcis-3177 56 13 network network NOUN fcis-3177 56 14 model model NOUN fcis-3177 56 15 [ [ X fcis-3177 56 16 10 10 NUM fcis-3177 56 17 ] ] PUNCT fcis-3177 56 18 , , PUNCT fcis-3177 56 19 which which PRON fcis-3177 56 20 had have VERB fcis-3177 56 21 strong strong ADJ fcis-3177 56 22 computational computational ADJ fcis-3177 56 23 power power NOUN fcis-3177 56 24 and and CCONJ fcis-3177 56 25 associative associative ADJ fcis-3177 56 26 memory memory NOUN fcis-3177 56 27 , , PUNCT fcis-3177 56 28 but but CCONJ fcis-3177 56 29 was be AUX fcis-3177 56 30 superseded supersede VERB fcis-3177 56 31 by by ADP fcis-3177 56 32 other other ADJ fcis-3177 56 33 artificial artificial ADJ fcis-3177 56 34 neural neural ADJ fcis-3177 56 35 networks network NOUN fcis-3177 56 36 and and CCONJ fcis-3177 56 37 traditional traditional ADJ fcis-3177 56 38 machine machine NOUN fcis-3177 56 39 learning learn VERB fcis-3177 56 40 algorithms algorithm NOUN fcis-3177 56 41 due due ADP fcis-3177 56 42 to to ADP fcis-3177 56 43 the the DET fcis-3177 56 44 difficulty difficulty NOUN fcis-3177 56 45 of of ADP fcis-3177 56 46 implementation implementation NOUN fcis-3177 56 47 . . PUNCT fcis-3177 57 1 jordan jordan PROPN fcis-3177 58 1 [ [ X fcis-3177 58 2 11 11 NUM fcis-3177 58 3 ] ] PUNCT fcis-3177 58 4 and and CCONJ fcis-3177 58 5 elman elman NOUN fcis-3177 59 1 [ [ X fcis-3177 59 2 12 12 NUM fcis-3177 59 3 ] ] PUNCT fcis-3177 59 4 proposed propose VERB fcis-3177 59 5 the the DET fcis-3177 59 6 recurrent recurrent ADJ fcis-3177 59 7 neural neural ADJ fcis-3177 59 8 network network NOUN fcis-3177 59 9 framework framework NOUN fcis-3177 59 10 in in ADP fcis-3177 59 11 1986 1986 NUM fcis-3177 59 12 and and CCONJ fcis-3177 59 13 the the DET fcis-3177 59 14 recurrent recurrent ADJ fcis-3177 59 15 neural neural ADJ fcis-3177 59 16 network network NOUN fcis-3177 59 17 framework framework NOUN fcis-3177 59 18 , , PUNCT fcis-3177 59 19 known know VERB fcis-3177 59 20 as as ADP fcis-3177 59 21 the the DET fcis-3177 59 22 simple simple ADJ fcis-3177 59 23 recurrent recurrent ADJ fcis-3177 59 24 network network NOUN fcis-3177 59 25 ( ( PUNCT fcis-3177 59 26 srn srn PROPN fcis-3177 59 27 ) ) PUNCT fcis-3177 59 28 , , PUNCT fcis-3177 59 29 was be AUX fcis-3177 59 30 proposed propose VERB fcis-3177 59 31 in in ADP fcis-3177 59 32 1990 1990 NUM fcis-3177 59 33 and and CCONJ fcis-3177 59 34 is be AUX fcis-3177 59 35 considered consider VERB fcis-3177 59 36 to to PART fcis-3177 59 37 be be AUX fcis-3177 59 38 the the DET fcis-3177 59 39 basic basic ADJ fcis-3177 59 40 version version NOUN fcis-3177 59 41 of of ADP fcis-3177 59 42 the the DET fcis-3177 59 43 current current ADJ fcis-3177 59 44 widely widely ADV fcis-3177 59 45 popular popular ADJ fcis-3177 59 46 rnn rnn NOUN fcis-3177 59 47 , , PUNCT fcis-3177 59 48 with with ADP fcis-3177 59 49 more more ADJ fcis-3177 59 50 complex complex ADJ fcis-3177 59 51 structures structure NOUN fcis-3177 59 52 that that PRON fcis-3177 59 53 have have AUX fcis-3177 59 54 since since SCONJ fcis-3177 59 55 emerged emerge VERB fcis-3177 59 56 being be AUX fcis-3177 59 57 considered consider VERB fcis-3177 59 58 as as ADP fcis-3177 59 59 variants variant NOUN fcis-3177 59 60 or or CCONJ fcis-3177 59 61 extensions extension NOUN fcis-3177 59 62 of of ADP fcis-3177 59 63 it it PRON fcis-3177 59 64 . . PUNCT fcis-3177 60 1 figure figure VERB fcis-3177 60 2 4 4 NUM fcis-3177 60 3 shows show VERB fcis-3177 60 4 the the DET fcis-3177 60 5 network network NOUN fcis-3177 60 6 structure structure NOUN fcis-3177 60 7 of of ADP fcis-3177 60 8 an an DET fcis-3177 60 9 rnn rnn NOUN fcis-3177 60 10 , , PUNCT fcis-3177 60 11 which which PRON fcis-3177 60 12 is be AUX fcis-3177 60 13 connected connect VERB fcis-3177 60 14 by by ADP fcis-3177 60 15 loops loop NOUN fcis-3177 60 16 on on ADP fcis-3177 60 17 the the DET fcis-3177 60 18 hidden hide VERB fcis-3177 60 19 layer layer NOUN fcis-3177 60 20 so so SCONJ fcis-3177 60 21 that that SCONJ fcis-3177 60 22 the the DET fcis-3177 60 23 network network NOUN fcis-3177 60 24 state state NOUN fcis-3177 60 25 at at ADP fcis-3177 60 26 the the DET fcis-3177 60 27 last last ADJ fcis-3177 60 28 moment moment NOUN fcis-3177 60 29 can can AUX fcis-3177 60 30 be be AUX fcis-3177 60 31 passed pass VERB fcis-3177 60 32 to to ADP fcis-3177 60 33 the the DET fcis-3177 60 34 current current ADJ fcis-3177 60 35 moment moment NOUN fcis-3177 60 36 and and CCONJ fcis-3177 60 37 the the DET fcis-3177 60 38 state state NOUN fcis-3177 60 39 at at ADP fcis-3177 60 40 the the DET fcis-3177 60 41 current current ADJ fcis-3177 60 42 moment moment NOUN fcis-3177 60 43 can can AUX fcis-3177 60 44 be be AUX fcis-3177 60 45 passed pass VERB fcis-3177 60 46 to to ADP fcis-3177 60 47 the the DET fcis-3177 60 48 next next ADJ fcis-3177 60 49 moment moment NOUN fcis-3177 60 50 . . PUNCT fcis-3177 61 1 figure figure VERB fcis-3177 61 2 4 4 NUM fcis-3177 61 3 . . PUNCT fcis-3177 61 4 basic basic ADJ fcis-3177 61 5 structure structure NOUN fcis-3177 61 6 of of ADP fcis-3177 61 7 recurrent recurrent ADJ fcis-3177 61 8 neural neural ADJ fcis-3177 61 9 network network NOUN fcis-3177 61 10 the the DET fcis-3177 61 11 rnn rnn NOUN fcis-3177 61 12 consists consist VERB fcis-3177 61 13 of of ADP fcis-3177 61 14 the the DET fcis-3177 61 15 input input NOUN fcis-3177 61 16 unit unit NOUN fcis-3177 61 17 xt xt PROPN fcis-3177 61 18 , , PUNCT fcis-3177 61 19 the the DET fcis-3177 61 20 output output NOUN fcis-3177 61 21 unit unit NOUN fcis-3177 61 22 ot ot INTJ fcis-3177 61 23 and and CCONJ fcis-3177 61 24 the the DET fcis-3177 61 25 hidden hidden ADJ fcis-3177 61 26 unit unit NOUN fcis-3177 61 27 ht ht PROPN fcis-3177 61 28 , , PUNCT fcis-3177 61 29 where where SCONJ fcis-3177 61 30 the the DET fcis-3177 61 31 ht ht PROPN fcis-3177 61 32 value value NOUN fcis-3177 61 33 depends depend VERB fcis-3177 61 34 not not PART fcis-3177 61 35 only only ADV fcis-3177 61 36 on on ADP fcis-3177 61 37 xt xt PROPN fcis-3177 62 1 but but CCONJ fcis-3177 62 2 also also ADV fcis-3177 62 3 on on ADP fcis-3177 62 4 ht-1 ht-1 X fcis-3177 62 5 . . PUNCT fcis-3177 63 1 w w PROPN fcis-3177 63 2 indicates indicate VERB fcis-3177 63 3 the the DET fcis-3177 63 4 weight weight NOUN fcis-3177 63 5 of of ADP fcis-3177 63 6 the the DET fcis-3177 63 7 input input NOUN fcis-3177 63 8 , , PUNCT fcis-3177 63 9 u u PRON fcis-3177 63 10 indicates indicate VERB fcis-3177 63 11 the the DET fcis-3177 63 12 weight weight NOUN fcis-3177 63 13 of of ADP fcis-3177 63 14 the the DET fcis-3177 63 15 input input NOUN fcis-3177 63 16 sample sample NOUN fcis-3177 63 17 at at ADP fcis-3177 63 18 the the DET fcis-3177 63 19 moment moment NOUN fcis-3177 63 20 , , PUNCT fcis-3177 63 21 and and CCONJ fcis-3177 63 22 v v NOUN fcis-3177 63 23 indicates indicate VERB fcis-3177 63 24 the the DET fcis-3177 63 25 weight weight NOUN fcis-3177 63 26 of of ADP fcis-3177 63 27 the the DET fcis-3177 63 28 output output NOUN fcis-3177 63 29 sample sample NOUN fcis-3177 63 30 . . PUNCT fcis-3177 64 1 in in ADP fcis-3177 64 2 this this DET fcis-3177 64 3 case case NOUN fcis-3177 64 4 , , PUNCT fcis-3177 64 5 f f PROPN fcis-3177 64 6 and and CCONJ fcis-3177 64 7 g g PROPN fcis-3177 64 8 are be AUX fcis-3177 64 9 both both PRON fcis-3177 64 10 activation activation NOUN fcis-3177 64 11 functions function NOUN fcis-3177 64 12 . . PUNCT fcis-3177 65 1 where where SCONJ fcis-3177 65 2 f f PROPN fcis-3177 65 3 can can AUX fcis-3177 65 4 be be AUX fcis-3177 65 5 an an DET fcis-3177 65 6 activation activation NOUN fcis-3177 65 7 function function NOUN fcis-3177 65 8 such such ADJ fcis-3177 65 9 as as ADP fcis-3177 65 10 tanh tanh NOUN fcis-3177 65 11 , , PUNCT fcis-3177 65 12 relu relu NOUN fcis-3177 65 13 , , PUNCT fcis-3177 65 14 sigmoid sigmoid NOUN fcis-3177 65 15 , , PUNCT fcis-3177 65 16 and and CCONJ fcis-3177 65 17 g g NOUN fcis-3177 65 18 is be AUX fcis-3177 65 19 usually usually ADV fcis-3177 65 20 softmax softmax ADJ fcis-3177 65 21 or or CCONJ fcis-3177 65 22 something something PRON fcis-3177 65 23 else else ADV fcis-3177 65 24 . . PUNCT fcis-3177 66 1 π‘₯𝑑 π‘₯𝑑 ADV fcis-3177 67 1 = = X fcis-3177 67 2 𝑔(𝑣 𝑔(𝑣 PROPN fcis-3177 67 3 βˆ™ βˆ™ PROPN fcis-3177 67 4 β„Ž β„Ž PROPN fcis-3177 67 5 𝑑 𝑑 NOUN fcis-3177 67 6 ) ) PUNCT fcis-3177 67 7 β„Ž β„Ž PROPN fcis-3177 67 8 𝑑 𝑑 PROPN fcis-3177 67 9 = = PROPN fcis-3177 67 10 𝑓(𝑒 𝑓(𝑒 PROPN fcis-3177 67 11 βˆ™ βˆ™ PROPN fcis-3177 67 12 π‘₯𝑑 π‘₯𝑑 ADV fcis-3177 67 13 + + NOUN fcis-3177 67 14 𝑀 𝑀 ADP fcis-3177 67 15 βˆ™ βˆ™ PROPN fcis-3177 67 16 β„Ž β„Ž PROPN fcis-3177 67 17 π‘‘βˆ’1 π‘‘βˆ’1 PROPN fcis-3177 67 18 ) ) PUNCT fcis-3177 67 19 3 3 X fcis-3177 67 20 . . X fcis-3177 68 1 few few ADJ fcis-3177 68 2 - - PUNCT fcis-3177 68 3 shot shot NOUN fcis-3177 68 4 learning learning NOUN fcis-3177 68 5 methods method NOUN fcis-3177 68 6 the the DET fcis-3177 68 7 development development NOUN fcis-3177 68 8 of of ADP fcis-3177 68 9 few few ADJ fcis-3177 68 10 - - PUNCT fcis-3177 68 11 shot shot NOUN fcis-3177 68 12 learning learning NOUN fcis-3177 68 13 has have AUX fcis-3177 68 14 so so ADV fcis-3177 68 15 far far ADV fcis-3177 68 16 received receive VERB fcis-3177 68 17 more more ADJ fcis-3177 68 18 and and CCONJ fcis-3177 68 19 more more ADJ fcis-3177 68 20 attention attention NOUN fcis-3177 68 21 from from ADP fcis-3177 68 22 scholars scholar NOUN fcis-3177 68 23 , , PUNCT fcis-3177 68 24 and and CCONJ fcis-3177 68 25 the the DET fcis-3177 68 26 current current ADJ fcis-3177 68 27 mainstream mainstream ADJ fcis-3177 68 28 few few ADJ fcis-3177 68 29 - - PUNCT fcis-3177 68 30 shot shot NOUN fcis-3177 68 31 learning learn VERB fcis-3177 68 32 methods method NOUN fcis-3177 68 33 based base VERB fcis-3177 68 34 on on ADP fcis-3177 68 35 the the DET fcis-3177 68 36 deep deep ADJ fcis-3177 68 37 neural neural ADJ fcis-3177 68 38 network network NOUN fcis-3177 68 39 are be AUX fcis-3177 68 40 classified classify VERB fcis-3177 68 41 into into ADP fcis-3177 68 42 four four NUM fcis-3177 68 43 major major ADJ fcis-3177 68 44 types type NOUN fcis-3177 68 45 : : PUNCT fcis-3177 68 46 data datum NOUN fcis-3177 68 47 augmentation augmentation NOUN fcis-3177 68 48 , , PUNCT fcis-3177 68 49 model model NOUN fcis-3177 68 50 fine fine ADV fcis-3177 68 51 - - PUNCT fcis-3177 68 52 tuning tuning NOUN fcis-3177 68 53 , , PUNCT fcis-3177 68 54 metric metric ADJ fcis-3177 68 55 learning learning NOUN fcis-3177 68 56 , , PUNCT fcis-3177 68 57 and and CCONJ fcis-3177 68 58 metalearning metalearning NOUN fcis-3177 68 59 . . PUNCT fcis-3177 69 1 3.1 3.1 NUM fcis-3177 69 2 . . PUNCT fcis-3177 70 1 data datum NOUN fcis-3177 70 2 augmentation augmentation VERB fcis-3177 70 3 the the DET fcis-3177 70 4 data data NOUN fcis-3177 70 5 augmentation augmentation NOUN fcis-3177 70 6 is be AUX fcis-3177 70 7 a a DET fcis-3177 70 8 technique technique NOUN fcis-3177 70 9 commonly commonly ADV fcis-3177 70 10 used use VERB fcis-3177 70 11 in in ADP fcis-3177 70 12 deep deep ADJ fcis-3177 70 13 learning learning NOUN fcis-3177 70 14 to to PART fcis-3177 70 15 expand expand VERB fcis-3177 70 16 the the DET fcis-3177 70 17 sample sample NOUN fcis-3177 70 18 size size NOUN fcis-3177 70 19 of of ADP fcis-3177 70 20 a a DET fcis-3177 70 21 training training NOUN fcis-3177 70 22 dataset dataset VERB fcis-3177 70 23 by by ADP fcis-3177 70 24 augmenting augment VERB fcis-3177 70 25 the the DET fcis-3177 70 26 data datum NOUN fcis-3177 70 27 with with ADP fcis-3177 70 28 prior prior ADJ fcis-3177 70 29 knowledge knowledge NOUN fcis-3177 70 30 and and CCONJ fcis-3177 70 31 increasing increase VERB fcis-3177 70 32 the the DET fcis-3177 70 33 diversity diversity NOUN fcis-3177 70 34 of of ADP fcis-3177 70 35 the the DET fcis-3177 70 36 data datum NOUN fcis-3177 70 37 to to PART fcis-3177 70 38 produce produce VERB fcis-3177 70 39 a a DET fcis-3177 70 40 larger large ADJ fcis-3177 70 41 dataset dataset NOUN fcis-3177 70 42 , , PUNCT fcis-3177 70 43 so so CCONJ fcis-3177 70 44 it it PRON fcis-3177 70 45 is be AUX fcis-3177 70 46 also also ADV fcis-3177 70 47 the the DET fcis-3177 70 48 most most ADV fcis-3177 70 49 direct direct ADJ fcis-3177 70 50 way way NOUN fcis-3177 70 51 to to PART fcis-3177 70 52 solve solve VERB fcis-3177 70 53 the the DET fcis-3177 70 54 problem problem NOUN fcis-3177 70 55 of of ADP fcis-3177 70 56 few few ADJ fcis-3177 70 57 - - PUNCT fcis-3177 70 58 shot shot NOUN fcis-3177 70 59 and and CCONJ fcis-3177 70 60 to to PART fcis-3177 70 61 avoid avoid VERB fcis-3177 70 62 overfitting overfitte VERB fcis-3177 70 63 the the DET fcis-3177 70 64 neural neural ADJ fcis-3177 70 65 network network NOUN fcis-3177 70 66 . . PUNCT fcis-3177 71 1 the the DET fcis-3177 71 2 early early ADJ fcis-3177 71 3 method method NOUN fcis-3177 71 4 of of ADP fcis-3177 71 5 data datum NOUN fcis-3177 71 6 enhancement enhancement NOUN fcis-3177 71 7 was be AUX fcis-3177 71 8 mainly mainly ADV fcis-3177 71 9 through through ADP fcis-3177 71 10 spatial spatial ADJ fcis-3177 71 11 transformations transformation NOUN fcis-3177 71 12 of of ADP fcis-3177 71 13 image image NOUN fcis-3177 71 14 data datum NOUN fcis-3177 71 15 , , PUNCT fcis-3177 71 16 including include VERB fcis-3177 71 17 rotating rotate VERB fcis-3177 71 18 , , PUNCT fcis-3177 71 19 cropping cropping NOUN fcis-3177 71 20 , , PUNCT fcis-3177 71 21 scaling scaling NOUN fcis-3177 71 22 , , PUNCT fcis-3177 71 23 panning pan VERB fcis-3177 71 24 , , PUNCT fcis-3177 71 25 adding add VERB fcis-3177 71 26 noise noise NOUN fcis-3177 71 27 , , PUNCT fcis-3177 71 28 and and CCONJ fcis-3177 71 29 changing change VERB fcis-3177 71 30 brightness brightness NOUN fcis-3177 71 31 or or CCONJ fcis-3177 71 32 contrast contrast NOUN fcis-3177 71 33 to to ADP fcis-3177 71 34 the the DET fcis-3177 71 35 image image NOUN fcis-3177 71 36 . . PUNCT fcis-3177 72 1 however however ADV fcis-3177 72 2 , , PUNCT fcis-3177 72 3 these these DET fcis-3177 72 4 methods method NOUN fcis-3177 72 5 are be AUX fcis-3177 72 6 highly highly ADV fcis-3177 72 7 dependent dependent ADJ fcis-3177 72 8 and and CCONJ fcis-3177 72 9 require require VERB fcis-3177 72 10 a a DET fcis-3177 72 11 lot lot NOUN fcis-3177 72 12 of of ADP fcis-3177 72 13 human human ADJ fcis-3177 72 14 resources resource NOUN fcis-3177 72 15 and and CCONJ fcis-3177 72 16 expertise expertise NOUN fcis-3177 72 17 . . PUNCT fcis-3177 73 1 in in ADP fcis-3177 73 2 addition addition NOUN fcis-3177 73 3 , , PUNCT fcis-3177 73 4 such such ADJ fcis-3177 73 5 methods method NOUN fcis-3177 73 6 are be AUX fcis-3177 73 7 less less ADV fcis-3177 73 8 transferable transferable ADJ fcis-3177 73 9 and and CCONJ fcis-3177 73 10 data datum NOUN fcis-3177 73 11 enhancement enhancement NOUN fcis-3177 73 12 methods method NOUN fcis-3177 73 13 developed develop VERB fcis-3177 73 14 for for ADP fcis-3177 73 15 one one NUM fcis-3177 73 16 particular particular ADJ fcis-3177 73 17 dataset dataset NOUN fcis-3177 73 18 are be AUX fcis-3177 73 19 challenging challenge VERB fcis-3177 73 20 to to PART fcis-3177 73 21 apply apply VERB fcis-3177 73 22 to to ADP fcis-3177 73 23 another another DET fcis-3177 73 24 dataset dataset NOUN fcis-3177 73 25 , , PUNCT fcis-3177 73 26 as as SCONJ fcis-3177 73 27 humans human NOUN fcis-3177 73 28 can can AUX fcis-3177 73 29 only only ADV fcis-3177 73 30 enumerate enumerate VERB fcis-3177 73 31 some some DET fcis-3177 73 32 possible possible ADJ fcis-3177 73 33 invariants invariant NOUN fcis-3177 73 34 and and CCONJ fcis-3177 73 35 better well ADJ fcis-3177 73 36 results result NOUN fcis-3177 73 37 can can AUX fcis-3177 73 38 only only ADV fcis-3177 73 39 be be AUX fcis-3177 73 40 obtained obtain VERB fcis-3177 73 41 when when SCONJ fcis-3177 73 42 applied apply VERB fcis-3177 73 43 to to ADP fcis-3177 73 44 certain certain ADJ fcis-3177 73 45 specific specific ADJ fcis-3177 73 46 datasets dataset NOUN fcis-3177 73 47 . . PUNCT fcis-3177 74 1 at at ADP fcis-3177 74 2 the the DET fcis-3177 74 3 same same ADJ fcis-3177 74 4 time time NOUN fcis-3177 74 5 , , PUNCT fcis-3177 74 6 such such ADJ fcis-3177 74 7 methods method NOUN fcis-3177 74 8 do do AUX fcis-3177 74 9 not not PART fcis-3177 74 10 address address VERB fcis-3177 74 11 the the DET fcis-3177 74 12 nature nature NOUN fcis-3177 74 13 of of ADP fcis-3177 74 14 few few ADJ fcis-3177 74 15 - - PUNCT fcis-3177 74 16 shot shot NOUN fcis-3177 74 17 learning learning NOUN fcis-3177 74 18 , , PUNCT fcis-3177 74 19 do do AUX fcis-3177 74 20 not not PART fcis-3177 74 21 enhance enhance VERB fcis-3177 74 22 the the DET fcis-3177 74 23 inference inference NOUN fcis-3177 74 24 and and CCONJ fcis-3177 74 25 generalization generalization NOUN fcis-3177 74 26 capabilities capability NOUN fcis-3177 74 27 of of ADP fcis-3177 74 28 the the DET fcis-3177 74 29 model model NOUN fcis-3177 74 30 , , PUNCT fcis-3177 74 31 and and CCONJ fcis-3177 74 32 do do AUX fcis-3177 74 33 not not PART fcis-3177 74 34 make make VERB fcis-3177 74 35 full full ADJ fcis-3177 74 36 use use NOUN fcis-3177 74 37 of of ADP fcis-3177 74 38 the the DET fcis-3177 74 39 information information NOUN fcis-3177 74 40 provided provide VERB fcis-3177 74 41 by by ADP fcis-3177 74 42 the the DET fcis-3177 74 43 base base NOUN fcis-3177 74 44 class class NOUN fcis-3177 74 45 dataset dataset NOUN fcis-3177 74 46 . . PUNCT fcis-3177 75 1 112 112 NUM fcis-3177 75 2 therefore therefore ADV fcis-3177 75 3 , , PUNCT fcis-3177 75 4 traditional traditional ADJ fcis-3177 75 5 artificially artificially ADV fcis-3177 75 6 enhanced enhance VERB fcis-3177 75 7 data data NOUN fcis-3177 75 8 methods method NOUN fcis-3177 75 9 can can AUX fcis-3177 75 10 not not PART fcis-3177 75 11 fully fully ADV fcis-3177 75 12 solve solve VERB fcis-3177 75 13 few few ADJ fcis-3177 75 14 - - PUNCT fcis-3177 75 15 shot shot NOUN fcis-3177 75 16 problems problem NOUN fcis-3177 75 17 . . PUNCT fcis-3177 76 1 the the DET fcis-3177 76 2 new new ADJ fcis-3177 76 3 data data NOUN fcis-3177 76 4 augmentation augmentation NOUN fcis-3177 76 5 method method NOUN fcis-3177 76 6 focuses focus VERB fcis-3177 76 7 on on ADP fcis-3177 76 8 generalizing generalize VERB fcis-3177 76 9 information information NOUN fcis-3177 76 10 between between ADP fcis-3177 76 11 similar similar ADJ fcis-3177 76 12 samples sample NOUN fcis-3177 76 13 from from ADP fcis-3177 76 14 a a DET fcis-3177 76 15 base base NOUN fcis-3177 76 16 class class NOUN fcis-3177 76 17 dataset dataset VERB fcis-3177 76 18 to to ADP fcis-3177 76 19 a a DET fcis-3177 76 20 new new ADJ fcis-3177 76 21 class class NOUN fcis-3177 76 22 of of ADP fcis-3177 76 23 few few ADJ fcis-3177 76 24 - - PUNCT fcis-3177 76 25 shot shot NOUN fcis-3177 76 26 . . PUNCT fcis-3177 77 1 the the DET fcis-3177 77 2 primary primary ADJ fcis-3177 77 3 approach approach NOUN fcis-3177 77 4 of of ADP fcis-3177 77 5 these these DET fcis-3177 77 6 methods method NOUN fcis-3177 77 7 is be AUX fcis-3177 77 8 to to PART fcis-3177 77 9 extend extend VERB fcis-3177 77 10 the the DET fcis-3177 77 11 training training NOUN fcis-3177 77 12 data datum NOUN fcis-3177 77 13 for for ADP fcis-3177 77 14 the the DET fcis-3177 77 15 new new ADJ fcis-3177 77 16 class class NOUN fcis-3177 77 17 of of ADP fcis-3177 77 18 few few ADJ fcis-3177 77 19 - - PUNCT fcis-3177 77 20 shot shot NOUN fcis-3177 77 21 by by ADP fcis-3177 77 22 learning learn VERB fcis-3177 77 23 a a DET fcis-3177 77 24 generative generative ADJ fcis-3177 77 25 model model NOUN fcis-3177 77 26 based base VERB fcis-3177 77 27 on on ADP fcis-3177 77 28 the the DET fcis-3177 77 29 base base NOUN fcis-3177 77 30 class class NOUN fcis-3177 77 31 dataset dataset NOUN fcis-3177 77 32 . . PUNCT fcis-3177 78 1 an an DET fcis-3177 78 2 extensive extensive ADJ fcis-3177 78 3 training training NOUN fcis-3177 78 4 data datum NOUN fcis-3177 78 5 set set VERB fcis-3177 78 6 can can AUX fcis-3177 78 7 typically typically ADV fcis-3177 78 8 be be AUX fcis-3177 78 9 generated generate VERB fcis-3177 78 10 with with ADP fcis-3177 78 11 only only ADV fcis-3177 78 12 one one NUM fcis-3177 78 13 or or CCONJ fcis-3177 78 14 a a DET fcis-3177 78 15 few few ADJ fcis-3177 78 16 training training NOUN fcis-3177 78 17 samples sample NOUN fcis-3177 78 18 . . PUNCT fcis-3177 79 1 for for ADP fcis-3177 79 2 example example NOUN fcis-3177 79 3 , , PUNCT fcis-3177 79 4 hariharan hariharan PROPN fcis-3177 79 5 et et PROPN fcis-3177 79 6 al al PROPN fcis-3177 79 7 . . PUNCT fcis-3177 80 1 [ [ X fcis-3177 80 2 14 14 NUM fcis-3177 80 3 ] ] PUNCT fcis-3177 80 4 argue argue VERB fcis-3177 80 5 that that SCONJ fcis-3177 80 6 the the DET fcis-3177 80 7 relationship relationship NOUN fcis-3177 80 8 between between ADP fcis-3177 80 9 images image NOUN fcis-3177 80 10 in in ADP fcis-3177 80 11 the the DET fcis-3177 80 12 same same ADJ fcis-3177 80 13 class class NOUN fcis-3177 80 14 is be AUX fcis-3177 80 15 shared share VERB fcis-3177 80 16 across across ADP fcis-3177 80 17 classes class NOUN fcis-3177 80 18 , , PUNCT fcis-3177 80 19 so so ADV fcis-3177 80 20 modeling model VERB fcis-3177 80 21 the the DET fcis-3177 80 22 relationship relationship NOUN fcis-3177 80 23 between between ADP fcis-3177 80 24 images image NOUN fcis-3177 80 25 in in ADP fcis-3177 80 26 the the DET fcis-3177 80 27 same same ADJ fcis-3177 80 28 class class NOUN fcis-3177 80 29 and and CCONJ fcis-3177 80 30 then then ADV fcis-3177 80 31 extending extend VERB fcis-3177 80 32 this this DET fcis-3177 80 33 relationship relationship NOUN fcis-3177 80 34 to to ADP fcis-3177 80 35 a a DET fcis-3177 80 36 new new ADJ fcis-3177 80 37 class class NOUN fcis-3177 80 38 of of ADP fcis-3177 80 39 few few ADJ fcis-3177 80 40 - - PUNCT fcis-3177 80 41 shot shot NOUN fcis-3177 80 42 generates generate VERB fcis-3177 80 43 some some DET fcis-3177 80 44 new new ADJ fcis-3177 80 45 training training NOUN fcis-3177 80 46 samples sample NOUN fcis-3177 80 47 , , PUNCT fcis-3177 80 48 while while SCONJ fcis-3177 80 49 wang wang PROPN fcis-3177 80 50 [ [ X fcis-3177 80 51 15 15 NUM fcis-3177 80 52 ] ] X fcis-3177 80 53 et et PROPN fcis-3177 80 54 al al PROPN fcis-3177 80 55 . . PUNCT fcis-3177 81 1 further further PROPN fcis-3177 81 2 couples couple NOUN fcis-3177 81 3 the the DET fcis-3177 81 4 generative generative ADJ fcis-3177 81 5 and and CCONJ fcis-3177 81 6 classification classification NOUN fcis-3177 81 7 networks network NOUN fcis-3177 81 8 together together ADV fcis-3177 81 9 end end NOUN fcis-3177 81 10 - - PUNCT fcis-3177 81 11 to to ADP fcis-3177 81 12 - - PUNCT fcis-3177 81 13 end end NOUN fcis-3177 81 14 to to PART fcis-3177 81 15 train train VERB fcis-3177 81 16 a a DET fcis-3177 81 17 neural neural ADJ fcis-3177 81 18 network network NOUN fcis-3177 81 19 adapted adapt VERB fcis-3177 81 20 to to ADP fcis-3177 81 21 few few ADJ fcis-3177 81 22 samples sample NOUN fcis-3177 81 23 neural neural ADJ fcis-3177 81 24 networks network NOUN fcis-3177 81 25 for for ADP fcis-3177 81 26 learning learn VERB fcis-3177 81 27 with with ADP fcis-3177 81 28 few few ADJ fcis-3177 81 29 - - PUNCT fcis-3177 81 30 shot shot NOUN fcis-3177 81 31 . . PUNCT fcis-3177 82 1 3.2 3.2 NUM fcis-3177 82 2 . . PUNCT fcis-3177 83 1 model model NOUN fcis-3177 83 2 fine fine ADV fcis-3177 83 3 - - PUNCT fcis-3177 83 4 tuning tune VERB fcis-3177 83 5 model model NOUN fcis-3177 83 6 fine fine ADV fcis-3177 83 7 - - PUNCT fcis-3177 83 8 tuning tuning NOUN fcis-3177 83 9 means mean NOUN fcis-3177 83 10 optimizing optimize VERB fcis-3177 83 11 the the DET fcis-3177 83 12 parameters parameter NOUN fcis-3177 83 13 of of ADP fcis-3177 83 14 a a DET fcis-3177 83 15 neural neural ADJ fcis-3177 83 16 network network NOUN fcis-3177 83 17 model model NOUN fcis-3177 83 18 that that PRON fcis-3177 83 19 has have AUX fcis-3177 83 20 been be AUX fcis-3177 83 21 pre pre VERB fcis-3177 83 22 - - VERB fcis-3177 83 23 trained train VERB fcis-3177 83 24 on on ADP fcis-3177 83 25 a a DET fcis-3177 83 26 significant significant ADJ fcis-3177 83 27 source source NOUN fcis-3177 83 28 dataset dataset NOUN fcis-3177 83 29 using use VERB fcis-3177 83 30 a a DET fcis-3177 83 31 corresponding corresponding ADJ fcis-3177 83 32 training training NOUN fcis-3177 83 33 strategy strategy NOUN fcis-3177 83 34 on on ADP fcis-3177 83 35 a a DET fcis-3177 83 36 target target NOUN fcis-3177 83 37 small small ADJ fcis-3177 83 38 sample sample NOUN fcis-3177 83 39 dataset dataset NOUN fcis-3177 83 40 . . PUNCT fcis-3177 84 1 the the DET fcis-3177 84 2 general general ADJ fcis-3177 84 3 approach approach NOUN fcis-3177 84 4 is be AUX fcis-3177 84 5 to to PART fcis-3177 84 6 pre pre VERB fcis-3177 84 7 - - VERB fcis-3177 84 8 train train VERB fcis-3177 84 9 the the DET fcis-3177 84 10 neural neural ADJ fcis-3177 84 11 network network NOUN fcis-3177 84 12 model model NOUN fcis-3177 84 13 on on ADP fcis-3177 84 14 the the DET fcis-3177 84 15 large large ADJ fcis-3177 84 16 - - PUNCT fcis-3177 84 17 scale scale NOUN fcis-3177 84 18 source source NOUN fcis-3177 84 19 dataset dataset NOUN fcis-3177 84 20 using use VERB fcis-3177 84 21 the the DET fcis-3177 84 22 general general ADJ fcis-3177 84 23 training training NOUN fcis-3177 84 24 strategy strategy NOUN fcis-3177 84 25 and and CCONJ fcis-3177 84 26 fix fix VERB fcis-3177 84 27 some some PRON fcis-3177 84 28 of of ADP fcis-3177 84 29 the the DET fcis-3177 84 30 parameters parameter NOUN fcis-3177 84 31 , , PUNCT fcis-3177 84 32 and and CCONJ fcis-3177 84 33 then then ADV fcis-3177 84 34 fine fine ADJ fcis-3177 84 35 - - PUNCT fcis-3177 84 36 tune tune NOUN fcis-3177 84 37 specific specific ADJ fcis-3177 84 38 parameters parameter NOUN fcis-3177 84 39 of of ADP fcis-3177 84 40 the the DET fcis-3177 84 41 neural neural ADJ fcis-3177 84 42 network network NOUN fcis-3177 84 43 model model NOUN fcis-3177 84 44 on on ADP fcis-3177 84 45 the the DET fcis-3177 84 46 small small ADJ fcis-3177 84 47 sample sample NOUN fcis-3177 84 48 dataset dataset VERB fcis-3177 84 49 to to PART fcis-3177 84 50 obtain obtain VERB fcis-3177 84 51 the the DET fcis-3177 84 52 target target NOUN fcis-3177 84 53 model model NOUN fcis-3177 84 54 . . PUNCT fcis-3177 85 1 figure figure NOUN fcis-3177 85 2 5 5 NUM fcis-3177 85 3 . . PUNCT fcis-3177 85 4 schematic schematic ADJ fcis-3177 85 5 diagram diagram NOUN fcis-3177 85 6 of of ADP fcis-3177 85 7 model model NOUN fcis-3177 85 8 fine fine ADV fcis-3177 85 9 - - PUNCT fcis-3177 85 10 tuning tune VERB fcis-3177 85 11 the the DET fcis-3177 85 12 parameters parameter NOUN fcis-3177 85 13 are be AUX fcis-3177 85 14 pre pre ADJ fcis-3177 85 15 - - VERB fcis-3177 85 16 trained train VERB fcis-3177 85 17 on on ADP fcis-3177 85 18 the the DET fcis-3177 85 19 source source NOUN fcis-3177 85 20 dataset dataset VERB fcis-3177 85 21 to to PART fcis-3177 85 22 help help VERB fcis-3177 85 23 the the DET fcis-3177 85 24 model model NOUN fcis-3177 85 25 converge converge VERB fcis-3177 85 26 quickly quickly ADV fcis-3177 85 27 on on ADP fcis-3177 85 28 the the DET fcis-3177 85 29 few few ADJ fcis-3177 85 30 - - PUNCT fcis-3177 85 31 shot shot NOUN fcis-3177 85 32 dataset dataset NOUN fcis-3177 85 33 . . PUNCT fcis-3177 86 1 suppose suppose VERB fcis-3177 86 2 the the DET fcis-3177 86 3 target target NOUN fcis-3177 86 4 dataset dataset VERB fcis-3177 86 5 and and CCONJ fcis-3177 86 6 the the DET fcis-3177 86 7 source source NOUN fcis-3177 86 8 dataset dataset NOUN fcis-3177 86 9 are be AUX fcis-3177 86 10 similarly similarly ADV fcis-3177 86 11 distributed distribute VERB fcis-3177 86 12 . . PUNCT fcis-3177 87 1 in in ADP fcis-3177 87 2 that that DET fcis-3177 87 3 case case NOUN fcis-3177 87 4 , , PUNCT fcis-3177 87 5 the the DET fcis-3177 87 6 top top ADJ fcis-3177 87 7 - - PUNCT fcis-3177 87 8 level level NOUN fcis-3177 87 9 features feature NOUN fcis-3177 87 10 of of ADP fcis-3177 87 11 the the DET fcis-3177 87 12 two two NUM fcis-3177 87 13 datasets dataset NOUN fcis-3177 87 14 are be AUX fcis-3177 87 15 highly highly ADV fcis-3177 87 16 similar similar ADJ fcis-3177 87 17 , , PUNCT fcis-3177 87 18 so so ADV fcis-3177 87 19 model model ADJ fcis-3177 87 20 fine fine ADV fcis-3177 87 21 - - PUNCT fcis-3177 87 22 tuning tuning NOUN fcis-3177 87 23 can can AUX fcis-3177 87 24 achieve achieve VERB fcis-3177 87 25 convergence convergence NOUN fcis-3177 87 26 and and CCONJ fcis-3177 87 27 generalization generalization NOUN fcis-3177 87 28 by by ADP fcis-3177 87 29 fine fine ADV fcis-3177 87 30 - - PUNCT fcis-3177 87 31 tuning tune VERB fcis-3177 87 32 the the DET fcis-3177 87 33 top top ADJ fcis-3177 87 34 - - PUNCT fcis-3177 87 35 level level NOUN fcis-3177 87 36 feature feature NOUN fcis-3177 87 37 extractor extractor NOUN fcis-3177 87 38 and and CCONJ fcis-3177 87 39 classifier classifier NOUN fcis-3177 87 40 only only ADV fcis-3177 87 41 . . PUNCT fcis-3177 88 1 model model NOUN fcis-3177 88 2 fine fine ADV fcis-3177 88 3 - - PUNCT fcis-3177 88 4 tuning tuning NOUN fcis-3177 88 5 requires require VERB fcis-3177 88 6 focusing focus VERB fcis-3177 88 7 on on ADP fcis-3177 88 8 architectural architectural ADJ fcis-3177 88 9 constraints constraint NOUN fcis-3177 88 10 , , PUNCT fcis-3177 88 11 the the DET fcis-3177 88 12 range range NOUN fcis-3177 88 13 of of ADP fcis-3177 88 14 parameters parameter NOUN fcis-3177 88 15 to to PART fcis-3177 88 16 be be AUX fcis-3177 88 17 tuned tune VERB fcis-3177 88 18 in in ADP fcis-3177 88 19 the the DET fcis-3177 88 20 network network NOUN fcis-3177 88 21 , , PUNCT fcis-3177 88 22 and and CCONJ fcis-3177 88 23 the the DET fcis-3177 88 24 learning learning NOUN fcis-3177 88 25 strategy strategy NOUN fcis-3177 88 26 . . PUNCT fcis-3177 89 1 howard howard PROPN fcis-3177 89 2 et et PROPN fcis-3177 89 3 al al PROPN fcis-3177 89 4 . . PUNCT fcis-3177 90 1 [ [ X fcis-3177 90 2 16 16 NUM fcis-3177 90 3 ] ] PUNCT fcis-3177 90 4 proposed propose VERB fcis-3177 90 5 the the DET fcis-3177 90 6 universal universal ADJ fcis-3177 90 7 language language NOUN fcis-3177 90 8 model model NOUN fcis-3177 90 9 fine fine ADV fcis-3177 90 10 - - PUNCT fcis-3177 90 11 tuning tuning NOUN fcis-3177 90 12 ( ( PUNCT fcis-3177 90 13 ulm ulm NOUN fcis-3177 90 14 -fit -fit VERB fcis-3177 90 15 ) ) PUNCT fcis-3177 90 16 for for ADP fcis-3177 90 17 text text NOUN fcis-3177 90 18 classification classification NOUN fcis-3177 90 19 in in ADP fcis-3177 90 20 2018 2018 NUM fcis-3177 90 21 . . PUNCT fcis-3177 91 1 the the DET fcis-3177 91 2 algorithm algorithm NOUN fcis-3177 91 3 fine fine NOUN fcis-3177 91 4 - - PUNCT fcis-3177 91 5 tunes tune NOUN fcis-3177 91 6 the the DET fcis-3177 91 7 language language NOUN fcis-3177 91 8 model model NOUN fcis-3177 91 9 in in ADP fcis-3177 91 10 terms term NOUN fcis-3177 91 11 of of ADP fcis-3177 91 12 changes change NOUN fcis-3177 91 13 in in ADP fcis-3177 91 14 the the DET fcis-3177 91 15 learning learning NOUN fcis-3177 91 16 rate rate NOUN fcis-3177 91 17 in in ADP fcis-3177 91 18 both both CCONJ fcis-3177 91 19 vertical vertical ADJ fcis-3177 91 20 and and CCONJ fcis-3177 91 21 horizontal horizontal ADJ fcis-3177 91 22 dimensions dimension NOUN fcis-3177 91 23 , , PUNCT fcis-3177 91 24 allowing allow VERB fcis-3177 91 25 the the DET fcis-3177 91 26 model model NOUN fcis-3177 91 27 to to PART fcis-3177 91 28 converge converge VERB fcis-3177 91 29 faster fast ADV fcis-3177 91 30 on on ADP fcis-3177 91 31 fewshot fewshot ADJ fcis-3177 91 32 datasets dataset NOUN fcis-3177 91 33 while while SCONJ fcis-3177 91 34 allowing allow VERB fcis-3177 91 35 the the DET fcis-3177 91 36 model model NOUN fcis-3177 91 37 to to PART fcis-3177 91 38 learn learn VERB fcis-3177 91 39 more more ADJ fcis-3177 91 40 knowledge knowledge NOUN fcis-3177 91 41 in in ADP fcis-3177 91 42 line line NOUN fcis-3177 91 43 with with ADP fcis-3177 91 44 the the DET fcis-3177 91 45 target target NOUN fcis-3177 91 46 task task NOUN fcis-3177 91 47 . . PUNCT fcis-3177 92 1 nakamura nakamura PROPN fcis-3177 92 2 et et PROPN fcis-3177 92 3 al al PROPN fcis-3177 92 4 . . PUNCT fcis-3177 93 1 [ [ X fcis-3177 93 2 17 17 NUM fcis-3177 93 3 ] ] PUNCT fcis-3177 93 4 proposed propose VERB fcis-3177 93 5 an an DET fcis-3177 93 6 approach approach NOUN fcis-3177 93 7 that that PRON fcis-3177 93 8 uses use VERB fcis-3177 93 9 a a DET fcis-3177 93 10 lower low ADJ fcis-3177 93 11 learning learning NOUN fcis-3177 93 12 rate rate NOUN fcis-3177 93 13 on on ADP fcis-3177 93 14 fewshot fewshot ADJ fcis-3177 93 15 datasets dataset NOUN fcis-3177 93 16 and and CCONJ fcis-3177 93 17 an an DET fcis-3177 93 18 adaptive adaptive ADJ fcis-3177 93 19 gradient gradient NOUN fcis-3177 93 20 optimizer optimizer NOUN fcis-3177 93 21 in in ADP fcis-3177 93 22 the the DET fcis-3177 93 23 finetuning finetuning NOUN fcis-3177 93 24 phase phase NOUN fcis-3177 93 25 . . PUNCT fcis-3177 94 1 the the DET fcis-3177 94 2 authors author NOUN fcis-3177 94 3 also also ADV fcis-3177 94 4 suggest suggest VERB fcis-3177 94 5 that that SCONJ fcis-3177 94 6 if if SCONJ fcis-3177 94 7 there there PRON fcis-3177 94 8 is be VERB fcis-3177 94 9 a a DET fcis-3177 94 10 large large ADJ fcis-3177 94 11 variability variability NOUN fcis-3177 94 12 between between ADP fcis-3177 94 13 the the DET fcis-3177 94 14 source source NOUN fcis-3177 94 15 and and CCONJ fcis-3177 94 16 target target NOUN fcis-3177 94 17 datasets dataset NOUN fcis-3177 94 18 , , PUNCT fcis-3177 94 19 the the DET fcis-3177 94 20 requirements requirement NOUN fcis-3177 94 21 can can AUX fcis-3177 94 22 be be AUX fcis-3177 94 23 achieved achieve VERB fcis-3177 94 24 by by ADP fcis-3177 94 25 tuning tune VERB fcis-3177 94 26 the the DET fcis-3177 94 27 entire entire ADJ fcis-3177 94 28 network network NOUN fcis-3177 94 29 . . PUNCT fcis-3177 95 1 the the DET fcis-3177 95 2 model model NOUN fcis-3177 95 3 fine fine ADV fcis-3177 95 4 - - PUNCT fcis-3177 95 5 tuning tuning NOUN fcis-3177 95 6 method method NOUN fcis-3177 95 7 is be AUX fcis-3177 95 8 simple simple ADJ fcis-3177 95 9 and and CCONJ fcis-3177 95 10 relies rely VERB fcis-3177 95 11 on on ADP fcis-3177 95 12 a a DET fcis-3177 95 13 small small ADJ fcis-3177 95 14 amount amount NOUN fcis-3177 95 15 of of ADP fcis-3177 95 16 data datum NOUN fcis-3177 95 17 , , PUNCT fcis-3177 95 18 requiring require VERB fcis-3177 95 19 only only ADV fcis-3177 95 20 a a DET fcis-3177 95 21 re re NOUN fcis-3177 95 22 - - NOUN fcis-3177 95 23 tuning tuning NOUN fcis-3177 95 24 of of ADP fcis-3177 95 25 the the DET fcis-3177 95 26 model model NOUN fcis-3177 95 27 parameters parameter NOUN fcis-3177 95 28 to to PART fcis-3177 95 29 achieve achieve VERB fcis-3177 95 30 the the DET fcis-3177 95 31 desired desire VERB fcis-3177 95 32 results result NOUN fcis-3177 95 33 quickly quickly ADV fcis-3177 95 34 . . PUNCT fcis-3177 96 1 however however ADV fcis-3177 96 2 , , PUNCT fcis-3177 96 3 there there PRON fcis-3177 96 4 are be VERB fcis-3177 96 5 limitations limitation NOUN fcis-3177 96 6 to to ADP fcis-3177 96 7 this this DET fcis-3177 96 8 approach approach NOUN fcis-3177 96 9 . . PUNCT fcis-3177 97 1 in in ADP fcis-3177 97 2 real real ADJ fcis-3177 97 3 - - PUNCT fcis-3177 97 4 life life NOUN fcis-3177 97 5 few few ADJ fcis-3177 97 6 - - PUNCT fcis-3177 97 7 shot shot NOUN fcis-3177 97 8 learning learn VERB fcis-3177 97 9 application application NOUN fcis-3177 97 10 scenarios scenario NOUN fcis-3177 97 11 , , PUNCT fcis-3177 97 12 the the DET fcis-3177 97 13 target target NOUN fcis-3177 97 14 dataset dataset VERB fcis-3177 97 15 and and CCONJ fcis-3177 97 16 the the DET fcis-3177 97 17 source source NOUN fcis-3177 97 18 dataset dataset NOUN fcis-3177 97 19 are be AUX fcis-3177 97 20 not not PART fcis-3177 97 21 necessarily necessarily ADV fcis-3177 97 22 similar similar ADJ fcis-3177 97 23 , , PUNCT fcis-3177 97 24 which which PRON fcis-3177 97 25 can can AUX fcis-3177 97 26 lead lead VERB fcis-3177 97 27 to to ADP fcis-3177 97 28 over over ADV fcis-3177 97 29 - - PUNCT fcis-3177 97 30 fitting fitting NOUN fcis-3177 97 31 of of ADP fcis-3177 97 32 the the DET fcis-3177 97 33 model model NOUN fcis-3177 97 34 on on ADP fcis-3177 97 35 the the DET fcis-3177 97 36 target target NOUN fcis-3177 97 37 dataset dataset NOUN fcis-3177 97 38 . . PUNCT fcis-3177 98 1 therefore therefore ADV fcis-3177 98 2 , , PUNCT fcis-3177 98 3 in in ADP fcis-3177 98 4 solving solve VERB fcis-3177 98 5 practical practical ADJ fcis-3177 98 6 problems problem NOUN fcis-3177 98 7 , , PUNCT fcis-3177 98 8 the the DET fcis-3177 98 9 model model NOUN fcis-3177 98 10 fine fine ADV fcis-3177 98 11 - - PUNCT fcis-3177 98 12 tuning tuning NOUN fcis-3177 98 13 method method NOUN fcis-3177 98 14 is be AUX fcis-3177 98 15 generally generally ADV fcis-3177 98 16 combined combine VERB fcis-3177 98 17 with with ADP fcis-3177 98 18 data datum NOUN fcis-3177 98 19 augmentation augmentation NOUN fcis-3177 98 20 , , PUNCT fcis-3177 98 21 metric metric ADJ fcis-3177 98 22 learning learning NOUN fcis-3177 98 23 , , PUNCT fcis-3177 98 24 or or CCONJ fcis-3177 98 25 meta meta ADV fcis-3177 98 26 - - PUNCT fcis-3177 98 27 learning learn VERB fcis-3177 98 28 methods method NOUN fcis-3177 98 29 to to PART fcis-3177 98 30 avoid avoid VERB fcis-3177 98 31 the the DET fcis-3177 98 32 problem problem NOUN fcis-3177 98 33 of of ADP fcis-3177 98 34 model model NOUN fcis-3177 98 35 overfitting overfitte VERB fcis-3177 98 36 due due ADP fcis-3177 98 37 to to ADP fcis-3177 98 38 small small ADJ fcis-3177 98 39 amounts amount NOUN fcis-3177 98 40 of of ADP fcis-3177 98 41 data datum NOUN fcis-3177 98 42 . . PUNCT fcis-3177 99 1 3.3 3.3 NUM fcis-3177 99 2 . . PUNCT fcis-3177 100 1 metric metric ADJ fcis-3177 100 2 learning learning NOUN fcis-3177 100 3 in in ADP fcis-3177 100 4 mathematical mathematical ADJ fcis-3177 100 5 concepts concept NOUN fcis-3177 100 6 , , PUNCT fcis-3177 100 7 the the DET fcis-3177 100 8 metric metric NOUN fcis-3177 100 9 is be AUX fcis-3177 100 10 a a DET fcis-3177 100 11 function function NOUN fcis-3177 100 12 that that PRON fcis-3177 100 13 measures measure VERB fcis-3177 100 14 the the DET fcis-3177 100 15 distance distance NOUN fcis-3177 100 16 between between ADP fcis-3177 100 17 two two NUM fcis-3177 100 18 elements element NOUN fcis-3177 100 19 , , PUNCT fcis-3177 100 20 also also ADV fcis-3177 100 21 called call VERB fcis-3177 100 22 a a DET fcis-3177 100 23 distance distance NOUN fcis-3177 100 24 function function NOUN fcis-3177 100 25 . . PUNCT fcis-3177 101 1 metric metric ADJ fcis-3177 101 2 learning learning NOUN fcis-3177 101 3 , , PUNCT fcis-3177 101 4 also also ADV fcis-3177 101 5 known know VERB fcis-3177 101 6 as as ADP fcis-3177 101 7 similarity similarity NOUN fcis-3177 101 8 learning learning NOUN fcis-3177 101 9 , , PUNCT fcis-3177 101 10 is be AUX fcis-3177 101 11 the the DET fcis-3177 101 12 calculation calculation NOUN fcis-3177 101 13 of of ADP fcis-3177 101 14 similarity similarity NOUN fcis-3177 101 15 between between ADP fcis-3177 101 16 two two NUM fcis-3177 101 17 samples sample NOUN fcis-3177 101 18 by by ADP fcis-3177 101 19 calculating calculate VERB fcis-3177 101 20 the the DET fcis-3177 101 21 distance distance NOUN fcis-3177 101 22 between between ADP fcis-3177 101 23 them they PRON fcis-3177 101 24 with with ADP fcis-3177 101 25 a a DET fcis-3177 101 26 given give VERB fcis-3177 101 27 distance distance NOUN fcis-3177 101 28 function function NOUN fcis-3177 101 29 [ [ X fcis-3177 101 30 18 18 NUM fcis-3177 101 31 ] ] PUNCT fcis-3177 101 32 , , PUNCT fcis-3177 101 33 commonly commonly ADV fcis-3177 101 34 used use VERB fcis-3177 101 35 distance distance NOUN fcis-3177 101 36 functions function NOUN fcis-3177 101 37 such such ADJ fcis-3177 101 38 as as ADP fcis-3177 101 39 euclidean euclidean ADJ fcis-3177 101 40 distance distance NOUN fcis-3177 101 41 , , PUNCT fcis-3177 101 42 mahala mahala PROPN fcis-3177 101 43 nobis nobis PROPN fcis-3177 101 44 distance distance NOUN fcis-3177 101 45 and and CCONJ fcis-3177 101 46 cosine cosine NOUN fcis-3177 101 47 similarity similarity NOUN fcis-3177 101 48 [ [ X fcis-3177 101 49 19 19 NUM fcis-3177 101 50 ] ] PUNCT fcis-3177 101 51 . . PUNCT fcis-3177 102 1 the the DET fcis-3177 102 2 metric metric ADV fcis-3177 102 3 - - PUNCT fcis-3177 102 4 based base VERB fcis-3177 102 5 learning learning NOUN fcis-3177 102 6 method method NOUN fcis-3177 102 7 for for ADP fcis-3177 102 8 a a DET fcis-3177 102 9 few few ADJ fcis-3177 102 10 samples sample NOUN fcis-3177 102 11 is be AUX fcis-3177 102 12 to to PART fcis-3177 102 13 determine determine VERB fcis-3177 102 14 the the DET fcis-3177 102 15 classification classification NOUN fcis-3177 102 16 result result NOUN fcis-3177 102 17 of of ADP fcis-3177 102 18 a a DET fcis-3177 102 19 sample sample NOUN fcis-3177 102 20 to to PART fcis-3177 102 21 be be AUX fcis-3177 102 22 classified classify VERB fcis-3177 102 23 by by ADP fcis-3177 102 24 calculating calculate VERB fcis-3177 102 25 the the DET fcis-3177 102 26 distance distance NOUN fcis-3177 102 27 between between ADP fcis-3177 102 28 the the DET fcis-3177 102 29 sample sample NOUN fcis-3177 102 30 to to PART fcis-3177 102 31 be be AUX fcis-3177 102 32 classified classify VERB fcis-3177 102 33 and and CCONJ fcis-3177 102 34 the the DET fcis-3177 102 35 known known ADJ fcis-3177 102 36 sample sample NOUN fcis-3177 102 37 to to PART fcis-3177 102 38 be be AUX fcis-3177 102 39 classified classify VERB fcis-3177 102 40 , , PUNCT fcis-3177 102 41 and and CCONJ fcis-3177 102 42 the the PRON fcis-3177 102 43 closer close ADJ fcis-3177 102 44 the the DET fcis-3177 102 45 distance distance NOUN fcis-3177 102 46 , , PUNCT fcis-3177 102 47 the the PRON fcis-3177 102 48 higher high ADJ fcis-3177 102 49 the the DET fcis-3177 102 50 similarity similarity NOUN fcis-3177 102 51 . . PUNCT fcis-3177 103 1 a a DET fcis-3177 103 2 significant significant ADJ fcis-3177 103 3 reason reason NOUN fcis-3177 103 4 general general ADJ fcis-3177 103 5 network network NOUN fcis-3177 103 6 models model NOUN fcis-3177 103 7 can can AUX fcis-3177 103 8 not not PART fcis-3177 103 9 be be AUX fcis-3177 103 10 adapted adapt VERB fcis-3177 103 11 to to ADP fcis-3177 103 12 the the DET fcis-3177 103 13 problem problem NOUN fcis-3177 103 14 of of ADP fcis-3177 103 15 few few ADJ fcis-3177 103 16 - - PUNCT fcis-3177 103 17 shot shot NOUN fcis-3177 103 18 learning learning NOUN fcis-3177 103 19 is be AUX fcis-3177 103 20 that that SCONJ fcis-3177 103 21 the the DET fcis-3177 103 22 number number NOUN fcis-3177 103 23 of of ADP fcis-3177 103 24 parameters parameter NOUN fcis-3177 103 25 to to PART fcis-3177 103 26 be be AUX fcis-3177 103 27 optimized optimize VERB fcis-3177 103 28 is be AUX fcis-3177 103 29 too too ADV fcis-3177 103 30 large large ADJ fcis-3177 103 31 , , PUNCT fcis-3177 103 32 and and CCONJ fcis-3177 103 33 using use VERB fcis-3177 103 34 only only ADV fcis-3177 103 35 the the DET fcis-3177 103 36 available available ADJ fcis-3177 103 37 data data NOUN fcis-3177 103 38 is be AUX fcis-3177 103 39 insufficient insufficient ADJ fcis-3177 103 40 to to PART fcis-3177 103 41 optimize optimize VERB fcis-3177 103 42 these these DET fcis-3177 103 43 parameters parameter NOUN fcis-3177 103 44 . . PUNCT fcis-3177 104 1 however however ADV fcis-3177 104 2 , , PUNCT fcis-3177 104 3 the the DET fcis-3177 104 4 metric metric ADJ fcis-3177 104 5 method method NOUN fcis-3177 104 6 , , PUNCT fcis-3177 104 7 as as ADP fcis-3177 104 8 a a DET fcis-3177 104 9 non non ADJ fcis-3177 104 10 - - ADJ fcis-3177 104 11 parametric parametric ADJ fcis-3177 104 12 method method NOUN fcis-3177 104 13 , , PUNCT fcis-3177 104 14 models model VERB fcis-3177 104 15 the the DET fcis-3177 104 16 distribution distribution NOUN fcis-3177 104 17 of of ADP fcis-3177 104 18 distances distance NOUN fcis-3177 104 19 between between ADP fcis-3177 104 20 samples sample NOUN fcis-3177 104 21 so so SCONJ fcis-3177 104 22 that that SCONJ fcis-3177 104 23 similar similar ADJ fcis-3177 104 24 samples sample NOUN fcis-3177 104 25 are be AUX fcis-3177 104 26 close close ADJ fcis-3177 104 27 together together ADV fcis-3177 104 28 and and CCONJ fcis-3177 104 29 different different ADJ fcis-3177 104 30 samples sample NOUN fcis-3177 104 31 are be AUX fcis-3177 104 32 separated separate VERB fcis-3177 104 33 . . PUNCT fcis-3177 105 1 at at ADP fcis-3177 105 2 the the DET fcis-3177 105 3 same same ADJ fcis-3177 105 4 time time NOUN fcis-3177 105 5 , , PUNCT fcis-3177 105 6 since since SCONJ fcis-3177 105 7 the the DET fcis-3177 105 8 metric metric ADJ fcis-3177 105 9 method method NOUN fcis-3177 105 10 measures measure VERB fcis-3177 105 11 the the DET fcis-3177 105 12 distance distance NOUN fcis-3177 105 13 between between ADP fcis-3177 105 14 samples sample NOUN fcis-3177 105 15 , , PUNCT fcis-3177 105 16 the the DET fcis-3177 105 17 model model NOUN fcis-3177 105 18 will will AUX fcis-3177 105 19 be be AUX fcis-3177 105 20 computationally computationally ADV fcis-3177 105 21 huge huge ADJ fcis-3177 105 22 for for ADP fcis-3177 105 23 training training NOUN fcis-3177 105 24 and and CCONJ fcis-3177 105 25 use use VERB fcis-3177 105 26 if if SCONJ fcis-3177 105 27 the the DET fcis-3177 105 28 number number NOUN fcis-3177 105 29 of of ADP fcis-3177 105 30 samples sample NOUN fcis-3177 105 31 is be AUX fcis-3177 105 32 large large ADJ fcis-3177 105 33 . . PUNCT fcis-3177 106 1 as as ADP fcis-3177 106 2 a a DET fcis-3177 106 3 result result NOUN fcis-3177 106 4 , , PUNCT fcis-3177 106 5 metric metric ADJ fcis-3177 106 6 methods method NOUN fcis-3177 106 7 have have AUX fcis-3177 106 8 been be AUX fcis-3177 106 9 widely widely ADV fcis-3177 106 10 used use VERB fcis-3177 106 11 in in ADP fcis-3177 106 12 recent recent ADJ fcis-3177 106 13 years year NOUN fcis-3177 106 14 in in ADP fcis-3177 106 15 the the DET fcis-3177 106 16 study study NOUN fcis-3177 106 17 of of ADP fcis-3177 106 18 few few ADJ fcis-3177 106 19 - - PUNCT fcis-3177 106 20 shot shot NOUN fcis-3177 106 21 learning learning NOUN fcis-3177 106 22 due due ADP fcis-3177 106 23 to to ADP fcis-3177 106 24 the the DET fcis-3177 106 25 minor minor ADJ fcis-3177 106 26 dataset dataset NOUN fcis-3177 106 27 nature nature NOUN fcis-3177 106 28 of of ADP fcis-3177 106 29 such such ADJ fcis-3177 106 30 learning learning NOUN fcis-3177 106 31 . . PUNCT fcis-3177 107 1 a a DET fcis-3177 107 2 common common ADJ fcis-3177 107 3 approach approach NOUN fcis-3177 107 4 is be AUX fcis-3177 107 5 to to PART fcis-3177 107 6 learn learn VERB fcis-3177 107 7 an an DET fcis-3177 107 8 end end VERB fcis-3177 107 9 - - PUNCT fcis-3177 107 10 to to ADP fcis-3177 107 11 - - PUNCT fcis-3177 107 12 end end NOUN fcis-3177 107 13 network network NOUN fcis-3177 107 14 to to PART fcis-3177 107 15 match match VERB fcis-3177 107 16 the the DET fcis-3177 107 17 data datum NOUN fcis-3177 107 18 distribution distribution NOUN fcis-3177 107 19 in in ADP fcis-3177 107 20 the the DET fcis-3177 107 21 representation representation NOUN fcis-3177 107 22 space space NOUN fcis-3177 107 23 and and CCONJ fcis-3177 107 24 the the DET fcis-3177 107 25 metric metric ADJ fcis-3177 107 26 function function NOUN fcis-3177 107 27 in in ADP fcis-3177 107 28 the the DET fcis-3177 107 29 upper upper ADJ fcis-3177 107 30 layer layer NOUN fcis-3177 107 31 , , PUNCT fcis-3177 107 32 e.g. e.g. ADV fcis-3177 107 33 , , PUNCT fcis-3177 107 34 koch koch PROPN fcis-3177 107 35 et et PROPN fcis-3177 107 36 al al PROPN fcis-3177 107 37 . . PUNCT fcis-3177 108 1 [ [ X fcis-3177 108 2 20 20 NUM fcis-3177 108 3 ] ] PUNCT fcis-3177 108 4 introduced introduce VERB fcis-3177 108 5 a a DET fcis-3177 108 6 two two NUM fcis-3177 108 7 - - PUNCT fcis-3177 108 8 way way NOUN fcis-3177 108 9 twin twin ADJ fcis-3177 108 10 network network NOUN fcis-3177 108 11 to to PART fcis-3177 108 12 learn learn VERB fcis-3177 108 13 the the DET fcis-3177 108 14 similarity similarity NOUN fcis-3177 108 15 between between ADP fcis-3177 108 16 two two NUM fcis-3177 108 17 images image NOUN fcis-3177 108 18 , , PUNCT fcis-3177 108 19 followed follow VERB fcis-3177 108 20 by by ADP fcis-3177 108 21 santoro santoro PROPN fcis-3177 108 22 et et PROPN fcis-3177 108 23 al al PROPN fcis-3177 108 24 . . PUNCT fcis-3177 109 1 [ [ X fcis-3177 109 2 21 21 NUM fcis-3177 109 3 ] ] PUNCT fcis-3177 109 4 who who PRON fcis-3177 109 5 proposed propose VERB fcis-3177 109 6 the the DET fcis-3177 109 7 memory memory NOUN fcis-3177 109 8 - - PUNCT fcis-3177 109 9 augmented augment VERB fcis-3177 109 10 neural neural ADJ fcis-3177 109 11 network network NOUN fcis-3177 109 12 ( ( PUNCT fcis-3177 109 13 mann mann PROPN fcis-3177 109 14 ) ) PUNCT fcis-3177 109 15 . . PUNCT fcis-3177 110 1 vinyals vinyal NOUN fcis-3177 110 2 et et PROPN fcis-3177 110 3 al al PROPN fcis-3177 110 4 . . PUNCT fcis-3177 111 1 [ [ X fcis-3177 111 2 22 22 NUM fcis-3177 111 3 ] ] PUNCT fcis-3177 111 4 proposed propose VERB fcis-3177 111 5 matching matching NOUN fcis-3177 111 6 network network NOUN fcis-3177 111 7 using use VERB fcis-3177 111 8 cosine cosine NOUN fcis-3177 111 9 similarity similarity NOUN fcis-3177 111 10 as as ADP fcis-3177 111 11 a a DET fcis-3177 111 12 metric metric ADJ fcis-3177 111 13 function function NOUN fcis-3177 111 14 , , PUNCT fcis-3177 111 15 snell snell PROPN fcis-3177 111 16 et et PROPN fcis-3177 111 17 al al PROPN fcis-3177 111 18 . . PUNCT fcis-3177 112 1 [ [ X fcis-3177 112 2 23 23 NUM fcis-3177 112 3 ] ] PUNCT fcis-3177 112 4 proposed propose VERB fcis-3177 112 5 prototypical prototypical ADJ fcis-3177 112 6 network network NOUN fcis-3177 112 7 using use VERB fcis-3177 112 8 euclidean euclidean ADJ fcis-3177 112 9 distance distance NOUN fcis-3177 112 10 as as ADP fcis-3177 112 11 a a DET fcis-3177 112 12 metric metric ADJ fcis-3177 112 13 function function NOUN fcis-3177 112 14 , , PUNCT fcis-3177 112 15 and and CCONJ fcis-3177 112 16 sung sing VERB fcis-3177 112 17 et et PROPN fcis-3177 112 18 al al PROPN fcis-3177 112 19 . . PUNCT fcis-3177 113 1 [ [ X fcis-3177 113 2 24 24 NUM fcis-3177 113 3 ] ] PUNCT fcis-3177 113 4 proposed propose VERB fcis-3177 113 5 relation relation NOUN fcis-3177 113 6 network network NOUN fcis-3177 113 7 ) ) PUNCT fcis-3177 113 8 using use VERB fcis-3177 113 9 a a DET fcis-3177 113 10 fully fully ADV fcis-3177 113 11 connected connected ADJ fcis-3177 113 12 network network NOUN fcis-3177 113 13 with with ADP fcis-3177 113 14 the the DET fcis-3177 113 15 representations representation NOUN fcis-3177 113 16 of of ADP fcis-3177 113 17 two two NUM fcis-3177 113 18 images image NOUN fcis-3177 113 19 stitched stitch VERB fcis-3177 113 20 together together ADV fcis-3177 113 21 as as ADP fcis-3177 113 22 input input NOUN fcis-3177 113 23 as as ADP fcis-3177 113 24 the the DET fcis-3177 113 25 metric metric ADJ fcis-3177 113 26 function function NOUN fcis-3177 113 27 . . PUNCT fcis-3177 114 1 taking take VERB fcis-3177 114 2 the the DET fcis-3177 114 3 classification classification NOUN fcis-3177 114 4 problem problem NOUN fcis-3177 114 5 as as ADP fcis-3177 114 6 an an DET fcis-3177 114 7 example example NOUN fcis-3177 114 8 , , PUNCT fcis-3177 114 9 the the DET fcis-3177 114 10 process process NOUN fcis-3177 114 11 of of ADP fcis-3177 114 12 metric metric ADJ fcis-3177 114 13 learning learn VERB fcis-3177 114 14 classification classification NOUN fcis-3177 114 15 of of ADP fcis-3177 114 16 a a DET fcis-3177 114 17 few few ADJ fcis-3177 114 18 samples sample NOUN fcis-3177 114 19 can can AUX fcis-3177 114 20 be be AUX fcis-3177 114 21 divided divide VERB fcis-3177 114 22 into into ADP fcis-3177 114 23 two two NUM fcis-3177 114 24 stages stage NOUN fcis-3177 114 25 : : PUNCT fcis-3177 114 26 mapping mapping NOUN fcis-3177 114 27 and and CCONJ fcis-3177 114 28 classification classification NOUN fcis-3177 114 29 . . PUNCT fcis-3177 115 1 as as SCONJ fcis-3177 115 2 shown show VERB fcis-3177 115 3 in in ADP fcis-3177 115 4 the the DET fcis-3177 115 5 figure figure NOUN fcis-3177 115 6 6 6 NUM fcis-3177 115 7 , , PUNCT fcis-3177 115 8 f f PROPN fcis-3177 115 9 is be AUX fcis-3177 115 10 the the DET fcis-3177 115 11 embedding embed VERB fcis-3177 115 12 model model NOUN fcis-3177 115 13 that that PRON fcis-3177 115 14 maps map VERB fcis-3177 115 15 the the DET fcis-3177 115 16 support support NOUN fcis-3177 115 17 set set VERB fcis-3177 115 18 samples sample NOUN fcis-3177 115 19 xj xj NOUN fcis-3177 115 20 to to ADP fcis-3177 115 21 the the DET fcis-3177 115 22 feature feature NOUN fcis-3177 115 23 space space NOUN fcis-3177 115 24 ; ; PUNCT fcis-3177 115 25 ΞΈf ΞΈf VERB fcis-3177 115 26 is be AUX fcis-3177 115 27 the the DET fcis-3177 115 28 parameter parameter NOUN fcis-3177 115 29 corresponding correspond VERB fcis-3177 115 30 to to ADP fcis-3177 115 31 f f PROPN fcis-3177 115 32 ; ; PUNCT fcis-3177 115 33 g g PROPN fcis-3177 115 34 is be AUX fcis-3177 115 35 the the DET fcis-3177 115 36 embedding embed VERB fcis-3177 115 37 model model NOUN fcis-3177 115 38 that that PRON fcis-3177 115 39 maps map VERB fcis-3177 115 40 the the DET fcis-3177 115 41 query query NOUN fcis-3177 115 42 set set VERB fcis-3177 115 43 samples sample NOUN fcis-3177 115 44 xi xi NOUN fcis-3177 115 45 to to ADP fcis-3177 115 46 the the DET fcis-3177 115 47 feature feature NOUN fcis-3177 115 48 space space NOUN fcis-3177 115 49 ; ; PUNCT fcis-3177 115 50 ΞΈg ΞΈg PRON fcis-3177 115 51 is be AUX fcis-3177 115 52 the the DET fcis-3177 115 53 parameter parameter NOUN fcis-3177 115 54 corresponding correspond VERB fcis-3177 115 55 to to ADP fcis-3177 115 56 g g NOUN fcis-3177 115 57 ; ; PUNCT fcis-3177 115 58 the the DET fcis-3177 115 59 metric metric ADJ fcis-3177 115 60 module module NOUN fcis-3177 115 61 is be AUX fcis-3177 115 62 used use VERB fcis-3177 115 63 to to PART fcis-3177 115 64 determine determine VERB fcis-3177 115 65 the the DET fcis-3177 115 66 sample sample NOUN fcis-3177 115 67 similarity similarity NOUN fcis-3177 115 68 between between ADP fcis-3177 115 69 the the DET fcis-3177 115 70 support support NOUN fcis-3177 115 71 set set VERB fcis-3177 115 72 and and CCONJ fcis-3177 115 73 the the DET fcis-3177 115 74 query query NOUN fcis-3177 115 75 set set NOUN fcis-3177 115 76 , , PUNCT fcis-3177 115 77 which which PRON fcis-3177 115 78 can can AUX fcis-3177 115 79 be be AUX fcis-3177 115 80 a a DET fcis-3177 115 81 simple simple ADJ fcis-3177 115 82 distance distance NOUN fcis-3177 115 83 metric metric ADJ fcis-3177 115 84 or or CCONJ fcis-3177 115 85 a a DET fcis-3177 115 86 learnability learnability NOUN fcis-3177 115 87 network network NOUN fcis-3177 115 88 . . PUNCT fcis-3177 116 1 finally finally ADV fcis-3177 116 2 , , PUNCT fcis-3177 116 3 the the DET fcis-3177 116 4 similarity similarity NOUN fcis-3177 116 5 output output NOUN fcis-3177 116 6 from from ADP fcis-3177 116 7 the the DET fcis-3177 116 8 metric metric ADJ fcis-3177 116 9 module module NOUN fcis-3177 116 10 is be AUX fcis-3177 116 11 used use VERB fcis-3177 116 12 to to PART fcis-3177 116 13 obtain obtain VERB fcis-3177 116 14 the the DET fcis-3177 116 15 classification classification NOUN fcis-3177 116 16 prediction prediction NOUN fcis-3177 116 17 results result NOUN fcis-3177 116 18 of of ADP fcis-3177 116 19 the the DET fcis-3177 116 20 query query NOUN fcis-3177 116 21 samples sample NOUN fcis-3177 116 22 . . PUNCT fcis-3177 117 1 113 113 NUM fcis-3177 117 2 figure figure NOUN fcis-3177 117 3 6 6 NUM fcis-3177 117 4 . . PUNCT fcis-3177 117 5 schematic schematic ADJ fcis-3177 117 6 diagram diagram NOUN fcis-3177 117 7 of of ADP fcis-3177 117 8 metric metric ADJ fcis-3177 117 9 learning learn VERB fcis-3177 117 10 the the DET fcis-3177 117 11 metric metric ADJ fcis-3177 117 12 learning learning NOUN fcis-3177 117 13 method method NOUN fcis-3177 117 14 reduces reduce VERB fcis-3177 117 15 the the DET fcis-3177 117 16 training training NOUN fcis-3177 117 17 difficulty difficulty NOUN fcis-3177 117 18 of of ADP fcis-3177 117 19 the the DET fcis-3177 117 20 feature feature NOUN fcis-3177 117 21 extractor extractor NOUN fcis-3177 117 22 with with ADP fcis-3177 117 23 the the DET fcis-3177 117 24 help help NOUN fcis-3177 117 25 of of ADP fcis-3177 117 26 a a DET fcis-3177 117 27 non non ADJ fcis-3177 117 28 - - ADJ fcis-3177 117 29 parametric parametric ADJ fcis-3177 117 30 classification classification NOUN fcis-3177 117 31 model model NOUN fcis-3177 117 32 , , PUNCT fcis-3177 117 33 which which PRON fcis-3177 117 34 is be AUX fcis-3177 117 35 more more ADV fcis-3177 117 36 suitable suitable ADJ fcis-3177 117 37 for for ADP fcis-3177 117 38 classification classification NOUN fcis-3177 117 39 with with ADP fcis-3177 117 40 fewer few ADJ fcis-3177 117 41 samples sample NOUN fcis-3177 117 42 , , PUNCT fcis-3177 117 43 and and CCONJ fcis-3177 117 44 the the DET fcis-3177 117 45 model model NOUN fcis-3177 117 46 structure structure NOUN fcis-3177 117 47 is be AUX fcis-3177 117 48 more more ADV fcis-3177 117 49 flexible flexible ADJ fcis-3177 117 50 and and CCONJ fcis-3177 117 51 efficient efficient ADJ fcis-3177 117 52 . . PUNCT fcis-3177 118 1 however however ADV fcis-3177 118 2 , , PUNCT fcis-3177 118 3 with with ADP fcis-3177 118 4 a a DET fcis-3177 118 5 small small ADJ fcis-3177 118 6 number number NOUN fcis-3177 118 7 of of ADP fcis-3177 118 8 samples sample NOUN fcis-3177 118 9 , , PUNCT fcis-3177 118 10 measuring measure VERB fcis-3177 118 11 similarity similarity NOUN fcis-3177 118 12 through through ADP fcis-3177 118 13 the the DET fcis-3177 118 14 traditional traditional ADJ fcis-3177 118 15 distance distance NOUN fcis-3177 118 16 function function NOUN fcis-3177 118 17 method method NOUN fcis-3177 118 18 can can AUX fcis-3177 118 19 lead lead VERB fcis-3177 118 20 to to ADP fcis-3177 118 21 a a DET fcis-3177 118 22 problematic problematic ADJ fcis-3177 118 23 improvement improvement NOUN fcis-3177 118 24 in in ADP fcis-3177 118 25 accuracy accuracy NOUN fcis-3177 118 26 . . PUNCT fcis-3177 119 1 3.4 3.4 NUM fcis-3177 119 2 . . PUNCT fcis-3177 120 1 meta meta VERB fcis-3177 120 2 - - PUNCT fcis-3177 120 3 learning learn VERB fcis-3177 120 4 the the DET fcis-3177 120 5 idea idea NOUN fcis-3177 120 6 of of ADP fcis-3177 120 7 meta meta ADV fcis-3177 120 8 - - PUNCT fcis-3177 120 9 learning learning NOUN fcis-3177 120 10 was be AUX fcis-3177 120 11 introduced introduce VERB fcis-3177 120 12 in in ADP fcis-3177 120 13 the the DET fcis-3177 120 14 1990s 1990 NOUN fcis-3177 120 15 [ [ X fcis-3177 120 16 25 25 NUM fcis-3177 120 17 ] ] PUNCT fcis-3177 120 18 with with ADP fcis-3177 120 19 the the DET fcis-3177 120 20 aim aim NOUN fcis-3177 120 21 of of ADP fcis-3177 120 22 building build VERB fcis-3177 120 23 a a DET fcis-3177 120 24 model model NOUN fcis-3177 120 25 that that PRON fcis-3177 120 26 can can AUX fcis-3177 120 27 learn learn VERB fcis-3177 120 28 new new ADJ fcis-3177 120 29 tasks task NOUN fcis-3177 120 30 quickly quickly ADV fcis-3177 120 31 , , PUNCT fcis-3177 120 32 while while SCONJ fcis-3177 120 33 the the DET fcis-3177 120 34 aim aim NOUN fcis-3177 120 35 of of ADP fcis-3177 120 36 few few ADJ fcis-3177 120 37 - - PUNCT fcis-3177 120 38 shot shot NOUN fcis-3177 120 39 learning learning NOUN fcis-3177 120 40 is be AUX fcis-3177 120 41 also also ADV fcis-3177 120 42 to to PART fcis-3177 120 43 gain gain VERB fcis-3177 120 44 the the DET fcis-3177 120 45 ability ability NOUN fcis-3177 120 46 to to PART fcis-3177 120 47 identify identify VERB fcis-3177 120 48 new new ADJ fcis-3177 120 49 categories category NOUN fcis-3177 120 50 from from ADP fcis-3177 120 51 a a DET fcis-3177 120 52 tiny tiny ADJ fcis-3177 120 53 number number NOUN fcis-3177 120 54 of of ADP fcis-3177 120 55 samples sample NOUN fcis-3177 120 56 . . PUNCT fcis-3177 121 1 in in ADP fcis-3177 121 2 terms term NOUN fcis-3177 121 3 of of ADP fcis-3177 121 4 task task NOUN fcis-3177 121 5 goals goal NOUN fcis-3177 121 6 , , PUNCT fcis-3177 121 7 the the DET fcis-3177 121 8 goals goal NOUN fcis-3177 121 9 of of ADP fcis-3177 121 10 meta meta NOUN fcis-3177 121 11 - - PUNCT fcis-3177 121 12 learning learn VERB fcis-3177 121 13 and and CCONJ fcis-3177 121 14 few few ADJ fcis-3177 121 15 - - PUNCT fcis-3177 121 16 shot shot NOUN fcis-3177 121 17 learning learning NOUN fcis-3177 121 18 are be AUX fcis-3177 121 19 the the DET fcis-3177 121 20 same same ADJ fcis-3177 121 21 . . PUNCT fcis-3177 122 1 with with ADP fcis-3177 122 2 the the DET fcis-3177 122 3 rapid rapid ADJ fcis-3177 122 4 development development NOUN fcis-3177 122 5 of of ADP fcis-3177 122 6 deep deep ADJ fcis-3177 122 7 learning learning NOUN fcis-3177 122 8 , , PUNCT fcis-3177 122 9 several several ADJ fcis-3177 122 10 researchers researcher NOUN fcis-3177 122 11 have have AUX fcis-3177 122 12 proposed propose VERB fcis-3177 122 13 using use VERB fcis-3177 122 14 meta meta ADJ fcis-3177 122 15 - - PUNCT fcis-3177 122 16 learning learn VERB fcis-3177 122 17 strategies strategy NOUN fcis-3177 122 18 to to PART fcis-3177 122 19 learn learn VERB fcis-3177 122 20 optimized optimize VERB fcis-3177 122 21 deep deep ADJ fcis-3177 122 22 learning learning NOUN fcis-3177 122 23 models model NOUN fcis-3177 123 1 [ [ X fcis-3177 123 2 26 26 NUM fcis-3177 123 3 ] ] PUNCT fcis-3177 123 4 . . PUNCT fcis-3177 124 1 although although SCONJ fcis-3177 124 2 the the DET fcis-3177 124 3 premise premise NOUN fcis-3177 124 4 of of ADP fcis-3177 124 5 a a DET fcis-3177 124 6 tiny tiny ADJ fcis-3177 124 7 sample sample NOUN fcis-3177 124 8 size size NOUN fcis-3177 124 9 may may AUX fcis-3177 124 10 seem seem VERB fcis-3177 124 11 contrary contrary ADV fcis-3177 124 12 to to ADP fcis-3177 124 13 deep deep ADJ fcis-3177 124 14 learning learning NOUN fcis-3177 124 15 , , PUNCT fcis-3177 124 16 we we PRON fcis-3177 124 17 can can AUX fcis-3177 124 18 still still ADV fcis-3177 124 19 achieve achieve VERB fcis-3177 124 20 few few ADJ fcis-3177 124 21 - - PUNCT fcis-3177 124 22 shot shot NOUN fcis-3177 124 23 learning learning NOUN fcis-3177 124 24 by by ADP fcis-3177 124 25 combining combine VERB fcis-3177 124 26 meta meta VERB fcis-3177 124 27 - - PUNCT fcis-3177 124 28 learning learning NOUN fcis-3177 124 29 and and CCONJ fcis-3177 124 30 deep deep ADJ fcis-3177 124 31 learning learning NOUN fcis-3177 124 32 . . PUNCT fcis-3177 125 1 meta meta VERB fcis-3177 125 2 - - PUNCT fcis-3177 125 3 learning learning NOUN fcis-3177 125 4 learns learn VERB fcis-3177 125 5 meta meta ADJ fcis-3177 125 6 - - PUNCT fcis-3177 125 7 knowledge knowledge NOUN fcis-3177 125 8 from from ADP fcis-3177 125 9 many many ADJ fcis-3177 125 10 a a DET fcis-3177 125 11 priori priori ADJ fcis-3177 125 12 tasks task NOUN fcis-3177 125 13 and and CCONJ fcis-3177 125 14 then then ADV fcis-3177 125 15 guides guide VERB fcis-3177 125 16 the the DET fcis-3177 125 17 model model NOUN fcis-3177 125 18 to to PART fcis-3177 125 19 perform perform VERB fcis-3177 125 20 better well ADV fcis-3177 125 21 on on ADP fcis-3177 125 22 few few ADJ fcis-3177 125 23 - - PUNCT fcis-3177 125 24 shot shot NOUN fcis-3177 125 25 tasks task NOUN fcis-3177 125 26 . . PUNCT fcis-3177 126 1 its its PRON fcis-3177 126 2 main main ADJ fcis-3177 126 3 idea idea NOUN fcis-3177 126 4 is be AUX fcis-3177 126 5 to to PART fcis-3177 126 6 devise devise VERB fcis-3177 126 7 a a DET fcis-3177 126 8 way way NOUN fcis-3177 126 9 to to PART fcis-3177 126 10 quickly quickly ADV fcis-3177 126 11 search search VERB fcis-3177 126 12 for for ADP fcis-3177 126 13 the the DET fcis-3177 126 14 model model NOUN fcis-3177 126 15 's 's PART fcis-3177 126 16 optimal optimal ADJ fcis-3177 126 17 parameters parameter NOUN fcis-3177 126 18 and and CCONJ fcis-3177 126 19 accelerate accelerate VERB fcis-3177 126 20 the the DET fcis-3177 126 21 learned learn VERB fcis-3177 126 22 model model NOUN fcis-3177 126 23 's 's PART fcis-3177 126 24 convergence convergence NOUN fcis-3177 126 25 on on ADP fcis-3177 126 26 a a DET fcis-3177 126 27 new new ADJ fcis-3177 126 28 task task NOUN fcis-3177 126 29 . . PUNCT fcis-3177 127 1 as as SCONJ fcis-3177 127 2 shown show VERB fcis-3177 127 3 in in ADP fcis-3177 127 4 figure figure NOUN fcis-3177 127 5 7 7 NUM fcis-3177 127 6 , , PUNCT fcis-3177 127 7 the the DET fcis-3177 127 8 object object NOUN fcis-3177 127 9 ( ( PUNCT fcis-3177 127 10 model model NOUN fcis-3177 127 11 ) ) PUNCT fcis-3177 127 12 being be AUX fcis-3177 127 13 optimized optimize VERB fcis-3177 127 14 is be AUX fcis-3177 127 15 referred refer VERB fcis-3177 127 16 to to ADP fcis-3177 127 17 as as ADP fcis-3177 127 18 the the DET fcis-3177 127 19 baselearner baselearner NOUN fcis-3177 127 20 , , PUNCT fcis-3177 127 21 and and CCONJ fcis-3177 127 22 the the DET fcis-3177 127 23 meta meta ADJ fcis-3177 127 24 - - PUNCT fcis-3177 127 25 learning learn VERB fcis-3177 127 26 process process NOUN fcis-3177 127 27 is be AUX fcis-3177 127 28 referred refer VERB fcis-3177 127 29 to to ADP fcis-3177 127 30 as as ADP fcis-3177 127 31 the the DET fcis-3177 127 32 meta meta ADJ fcis-3177 127 33 - - PUNCT fcis-3177 127 34 learner learner NOUN fcis-3177 127 35 . . PUNCT fcis-3177 128 1 the the DET fcis-3177 128 2 base base NOUN fcis-3177 128 3 - - PUNCT fcis-3177 128 4 learner learner NOUN fcis-3177 128 5 ( ( PUNCT fcis-3177 128 6 model model NOUN fcis-3177 128 7 ) ) PUNCT fcis-3177 128 8 goal goal NOUN fcis-3177 128 9 is be AUX fcis-3177 128 10 to to PART fcis-3177 128 11 quickly quickly ADV fcis-3177 128 12 learn learn VERB fcis-3177 128 13 a a DET fcis-3177 128 14 new new ADJ fcis-3177 128 15 task task NOUN fcis-3177 128 16 using use VERB fcis-3177 128 17 a a DET fcis-3177 128 18 small small ADJ fcis-3177 128 19 amount amount NOUN fcis-3177 128 20 of of ADP fcis-3177 128 21 data datum NOUN fcis-3177 128 22 . . PUNCT fcis-3177 129 1 therefore therefore ADV fcis-3177 129 2 , , PUNCT fcis-3177 129 3 the the DET fcis-3177 129 4 base base NOUN fcis-3177 129 5 - - PUNCT fcis-3177 129 6 learner learner NOUN fcis-3177 129 7 is be AUX fcis-3177 129 8 also also ADV fcis-3177 129 9 known know VERB fcis-3177 129 10 as as ADP fcis-3177 129 11 the the DET fcis-3177 129 12 fast fast ADJ fcis-3177 129 13 - - PUNCT fcis-3177 129 14 learner learner NOUN fcis-3177 129 15 . . PUNCT fcis-3177 130 1 the the DET fcis-3177 130 2 goal goal NOUN fcis-3177 130 3 of of ADP fcis-3177 130 4 the the DET fcis-3177 130 5 meta meta ADJ fcis-3177 130 6 - - PUNCT fcis-3177 130 7 learner learner NOUN fcis-3177 130 8 is be AUX fcis-3177 130 9 to to PART fcis-3177 130 10 train train VERB fcis-3177 130 11 the the DET fcis-3177 130 12 base base NOUN fcis-3177 130 13 - - PUNCT fcis-3177 130 14 learner learner NOUN fcis-3177 130 15 with with ADP fcis-3177 130 16 many many ADJ fcis-3177 130 17 different different ADJ fcis-3177 130 18 learning learning NOUN fcis-3177 130 19 tasks task NOUN fcis-3177 130 20 so so SCONJ fcis-3177 130 21 that that SCONJ fcis-3177 130 22 the the DET fcis-3177 130 23 trained train VERB fcis-3177 130 24 base base NOUN fcis-3177 130 25 - - PUNCT fcis-3177 130 26 learner learner NOUN fcis-3177 130 27 can can AUX fcis-3177 130 28 quickly quickly ADV fcis-3177 130 29 learn learn VERB fcis-3177 130 30 a a DET fcis-3177 130 31 new new ADJ fcis-3177 130 32 task task NOUN fcis-3177 130 33 using use VERB fcis-3177 130 34 only only ADV fcis-3177 130 35 a a DET fcis-3177 130 36 small small ADJ fcis-3177 130 37 number number NOUN fcis-3177 130 38 of of ADP fcis-3177 130 39 training training NOUN fcis-3177 130 40 samples sample NOUN fcis-3177 130 41 , , PUNCT fcis-3177 130 42 that that PRON fcis-3177 130 43 is be AUX fcis-3177 130 44 the the DET fcis-3177 130 45 few few ADJ fcis-3177 130 46 - - PUNCT fcis-3177 130 47 shot shot NOUN fcis-3177 130 48 learning learning NOUN fcis-3177 130 49 task task NOUN fcis-3177 130 50 . . PUNCT fcis-3177 131 1 the the DET fcis-3177 131 2 meta meta ADV fcis-3177 131 3 - - PUNCT fcis-3177 131 4 learning learning NOUN fcis-3177 131 5 based base VERB fcis-3177 131 6 few few ADJ fcis-3177 131 7 - - PUNCT fcis-3177 131 8 shot shot NOUN fcis-3177 131 9 learning learning NOUN fcis-3177 131 10 method method NOUN fcis-3177 131 11 learns learn VERB fcis-3177 131 12 a a DET fcis-3177 131 13 priori priori ADJ fcis-3177 131 14 tasks task NOUN fcis-3177 131 15 through through ADP fcis-3177 131 16 the the DET fcis-3177 131 17 base base NOUN fcis-3177 131 18 - - PUNCT fcis-3177 131 19 learner learner NOUN fcis-3177 131 20 , , PUNCT fcis-3177 131 21 giving give VERB fcis-3177 131 22 the the DET fcis-3177 131 23 model model NOUN fcis-3177 131 24 the the DET fcis-3177 131 25 ability ability NOUN fcis-3177 131 26 to to PART fcis-3177 131 27 learn learn VERB fcis-3177 131 28 automatically automatically ADV fcis-3177 131 29 , , PUNCT fcis-3177 131 30 to to PART fcis-3177 131 31 learn learn VERB fcis-3177 131 32 beyond beyond ADP fcis-3177 131 33 training training NOUN fcis-3177 131 34 and and CCONJ fcis-3177 131 35 to to PART fcis-3177 131 36 become become VERB fcis-3177 131 37 flexible flexible ADJ fcis-3177 131 38 in in ADP fcis-3177 131 39 solving solve VERB fcis-3177 131 40 different different ADJ fcis-3177 131 41 categories category NOUN fcis-3177 131 42 of of ADP fcis-3177 131 43 problems problem NOUN fcis-3177 131 44 . . PUNCT fcis-3177 132 1 figure figure VERB fcis-3177 132 2 7 7 NUM fcis-3177 132 3 . . PUNCT fcis-3177 132 4 schematic schematic ADJ fcis-3177 132 5 diagram diagram NOUN fcis-3177 132 6 of of ADP fcis-3177 132 7 meta meta ADJ fcis-3177 132 8 - - PUNCT fcis-3177 132 9 learning learn VERB fcis-3177 132 10 models model NOUN fcis-3177 132 11 based base VERB fcis-3177 132 12 on on ADP fcis-3177 132 13 the the DET fcis-3177 132 14 meta meta ADV fcis-3177 132 15 - - PUNCT fcis-3177 132 16 learning learn VERB fcis-3177 132 17 method method NOUN fcis-3177 132 18 are be AUX fcis-3177 132 19 more more ADV fcis-3177 132 20 complex complex ADJ fcis-3177 132 21 and and CCONJ fcis-3177 132 22 require require VERB fcis-3177 132 23 more more ADJ fcis-3177 132 24 improvement improvement NOUN fcis-3177 132 25 aspects aspect NOUN fcis-3177 132 26 . . PUNCT fcis-3177 133 1 for for ADP fcis-3177 133 2 example example NOUN fcis-3177 133 3 , , PUNCT fcis-3177 133 4 how how SCONJ fcis-3177 133 5 to to PART fcis-3177 133 6 set set VERB fcis-3177 133 7 task task NOUN fcis-3177 133 8 - - PUNCT fcis-3177 133 9 general general ADJ fcis-3177 133 10 and and CCONJ fcis-3177 133 11 task task NOUN fcis-3177 133 12 - - PUNCT fcis-3177 133 13 specific specific ADJ fcis-3177 133 14 parameters parameter NOUN fcis-3177 133 15 and and CCONJ fcis-3177 133 16 effectively effectively ADV fcis-3177 133 17 train train VERB fcis-3177 133 18 meta meta ADJ fcis-3177 133 19 - - PUNCT fcis-3177 133 20 learning learn VERB fcis-3177 133 21 models model NOUN fcis-3177 133 22 have have AUX fcis-3177 133 23 been be AUX fcis-3177 133 24 the the DET fcis-3177 133 25 hotspot hotspot NOUN fcis-3177 133 26 of of ADP fcis-3177 133 27 research research NOUN fcis-3177 133 28 in in ADP fcis-3177 133 29 this this DET fcis-3177 133 30 field field NOUN fcis-3177 133 31 . . PUNCT fcis-3177 134 1 in in ADP fcis-3177 134 2 addition addition NOUN fcis-3177 134 3 , , PUNCT fcis-3177 134 4 the the DET fcis-3177 134 5 data datum NOUN fcis-3177 134 6 for for ADP fcis-3177 134 7 different different ADJ fcis-3177 134 8 tasks task NOUN fcis-3177 134 9 have have VERB fcis-3177 134 10 different different ADJ fcis-3177 134 11 distributions distribution NOUN fcis-3177 134 12 , , PUNCT fcis-3177 134 13 and and CCONJ fcis-3177 134 14 significant significant ADJ fcis-3177 134 15 differences difference NOUN fcis-3177 134 16 in in ADP fcis-3177 134 17 data datum NOUN fcis-3177 134 18 distribution distribution NOUN fcis-3177 134 19 can can AUX fcis-3177 134 20 lead lead VERB fcis-3177 134 21 to to ADP fcis-3177 134 22 difficulties difficulty NOUN fcis-3177 134 23 in in ADP fcis-3177 134 24 the the DET fcis-3177 134 25 convergence convergence NOUN fcis-3177 134 26 of of ADP fcis-3177 134 27 the the DET fcis-3177 134 28 model model NOUN fcis-3177 134 29 . . PUNCT fcis-3177 135 1 4 4 X fcis-3177 135 2 . . X fcis-3177 135 3 datasets dataset NOUN fcis-3177 135 4 for for ADP fcis-3177 135 5 few few ADJ fcis-3177 135 6 - - PUNCT fcis-3177 135 7 shot shot NOUN fcis-3177 135 8 learning learn VERB fcis-3177 135 9 the the DET fcis-3177 135 10 research research NOUN fcis-3177 135 11 on on ADP fcis-3177 135 12 few few ADJ fcis-3177 135 13 - - PUNCT fcis-3177 135 14 shot shot NOUN fcis-3177 135 15 learning learning NOUN fcis-3177 135 16 mainly mainly ADV fcis-3177 135 17 focused focus VERB fcis-3177 135 18 on on ADP fcis-3177 135 19 image image NOUN fcis-3177 135 20 classification classification NOUN fcis-3177 135 21 and and CCONJ fcis-3177 135 22 recognition recognition NOUN fcis-3177 135 23 tasks task NOUN fcis-3177 135 24 . . PUNCT fcis-3177 136 1 the the DET fcis-3177 136 2 most most ADV fcis-3177 136 3 commonly commonly ADV fcis-3177 136 4 used use VERB fcis-3177 136 5 dataset dataset NOUN fcis-3177 136 6 for for ADP fcis-3177 136 7 one one NUM fcis-3177 136 8 - - PUNCT fcis-3177 136 9 shot shot NOUN fcis-3177 136 10 learning learning NOUN fcis-3177 136 11 is be AUX fcis-3177 136 12 the the DET fcis-3177 136 13 omniglot omniglot NOUN fcis-3177 136 14 dataset dataset NOUN fcis-3177 136 15 , , PUNCT fcis-3177 136 16 and and CCONJ fcis-3177 136 17 the the DET fcis-3177 136 18 most most ADV fcis-3177 136 19 commonly commonly ADV fcis-3177 136 20 used use VERB fcis-3177 136 21 dataset dataset NOUN fcis-3177 136 22 for for ADP fcis-3177 136 23 few few ADJ fcis-3177 136 24 - - PUNCT fcis-3177 136 25 shot shot NOUN fcis-3177 136 26 learning learning NOUN fcis-3177 136 27 is be AUX fcis-3177 136 28 mini mini NOUN fcis-3177 136 29 - - NOUN fcis-3177 136 30 imagenet imagenet ADJ fcis-3177 136 31 , , PUNCT fcis-3177 136 32 in in ADP fcis-3177 136 33 addition addition NOUN fcis-3177 136 34 to to ADP fcis-3177 136 35 tiered tiere VERB fcis-3177 136 36 - - PUNCT fcis-3177 136 37 imagenet imagenet NOUN fcis-3177 136 38 and and CCONJ fcis-3177 136 39 cub-200 cub-200 VERB fcis-3177 136 40 . . PUNCT fcis-3177 137 1 as as ADV fcis-3177 137 2 well well ADV fcis-3177 137 3 as as ADP fcis-3177 137 4 cifar-100 cifar-100 NOUN fcis-3177 137 5 , , PUNCT fcis-3177 138 1 stanford stanford PROPN fcis-3177 138 2 dogs dog NOUN fcis-3177 138 3 and and CCONJ fcis-3177 138 4 stanford stanford PROPN fcis-3177 138 5 cars car NOUN fcis-3177 138 6 , , PUNCT fcis-3177 138 7 which which PRON fcis-3177 138 8 are be AUX fcis-3177 138 9 the the DET fcis-3177 138 10 most most ADV fcis-3177 138 11 commonly commonly ADV fcis-3177 138 12 used use VERB fcis-3177 138 13 datasets dataset NOUN fcis-3177 138 14 for for ADP fcis-3177 138 15 fine fine ADV fcis-3177 138 16 - - PUNCT fcis-3177 138 17 grained grain VERB fcis-3177 138 18 few few ADJ fcis-3177 138 19 - - PUNCT fcis-3177 138 20 shot shot NOUN fcis-3177 138 21 image image NOUN fcis-3177 138 22 classification classification NOUN fcis-3177 138 23 . . PUNCT fcis-3177 139 1 in in ADP fcis-3177 139 2 recent recent ADJ fcis-3177 139 3 years year NOUN fcis-3177 139 4 , , PUNCT fcis-3177 139 5 the the DET fcis-3177 139 6 field field NOUN fcis-3177 139 7 of of ADP fcis-3177 139 8 natural natural ADJ fcis-3177 139 9 language language NOUN fcis-3177 139 10 processing processing NOUN fcis-3177 139 11 has have AUX fcis-3177 139 12 also also ADV fcis-3177 139 13 started start VERB fcis-3177 139 14 to to PART fcis-3177 139 15 see see VERB fcis-3177 139 16 the the DET fcis-3177 139 17 emergence emergence NOUN fcis-3177 139 18 of of ADP fcis-3177 139 19 few few ADJ fcis-3177 139 20 - - PUNCT fcis-3177 139 21 shot shot NOUN fcis-3177 139 22 learning learn VERB fcis-3177 139 23 datasets dataset NOUN fcis-3177 139 24 , , PUNCT fcis-3177 139 25 and and CCONJ fcis-3177 139 26 the the DET fcis-3177 139 27 most most ADV fcis-3177 139 28 common common ADJ fcis-3177 139 29 ones one NOUN fcis-3177 139 30 are be AUX fcis-3177 139 31 fewrel fewrel NOUN fcis-3177 139 32 , , PUNCT fcis-3177 139 33 arsc arsc ADJ fcis-3177 139 34 and and CCONJ fcis-3177 139 35 odic odic ADJ fcis-3177 139 36 datasets dataset NOUN fcis-3177 139 37 . . PUNCT fcis-3177 140 1 figure figure NOUN fcis-3177 140 2 8 8 NUM fcis-3177 140 3 . . PUNCT fcis-3177 141 1 sample sample NOUN fcis-3177 141 2 image image NOUN fcis-3177 141 3 of of ADP fcis-3177 141 4 mini mini ADJ fcis-3177 141 5 - - NOUN fcis-3177 141 6 imagenet imagenet NOUN fcis-3177 141 7 dataset dataset VERB fcis-3177 141 8 4.1 4.1 NUM fcis-3177 141 9 . . PUNCT fcis-3177 142 1 computer computer NOUN fcis-3177 142 2 vision vision NOUN fcis-3177 142 3 field field PROPN fcis-3177 142 4 1 1 NUM fcis-3177 142 5 ) ) PUNCT fcis-3177 142 6 . . PUNCT fcis-3177 143 1 the the DET fcis-3177 143 2 mini mini ADJ fcis-3177 143 3 - - ADJ fcis-3177 143 4 imagenet imagenet ADJ fcis-3177 143 5 dataset dataset NOUN fcis-3177 143 6 [ [ X fcis-3177 143 7 27 27 NUM fcis-3177 143 8 ] ] PUNCT fcis-3177 143 9 was be AUX fcis-3177 143 10 extracted extract VERB fcis-3177 143 11 from from ADP fcis-3177 143 12 imagenet imagenet NOUN fcis-3177 143 13 by by ADP fcis-3177 143 14 the the DET fcis-3177 143 15 deepmind deepmind NOUN fcis-3177 143 16 and and CCONJ fcis-3177 143 17 is be AUX fcis-3177 143 18 used use VERB fcis-3177 143 19 explicitly explicitly ADV fcis-3177 143 20 for for ADP fcis-3177 143 21 fewshot fewshot ADJ fcis-3177 143 22 learning learning NOUN fcis-3177 143 23 research research NOUN fcis-3177 143 24 , , PUNCT fcis-3177 143 25 and and CCONJ fcis-3177 143 26 is be AUX fcis-3177 143 27 a a DET fcis-3177 143 28 benchmark benchmark NOUN fcis-3177 143 29 dataset dataset NOUN fcis-3177 143 30 in in ADP fcis-3177 143 31 the the DET fcis-3177 143 32 field field NOUN fcis-3177 143 33 of of ADP fcis-3177 143 34 meta meta NOUN fcis-3177 143 35 - - PUNCT fcis-3177 143 36 learning learn VERB fcis-3177 143 37 and and CCONJ fcis-3177 143 38 few few ADJ fcis-3177 143 39 - - PUNCT fcis-3177 143 40 shot shot NOUN fcis-3177 143 41 . . PUNCT fcis-3177 144 1 the the DET fcis-3177 144 2 training training NOUN fcis-3177 144 3 and and CCONJ fcis-3177 144 4 test test NOUN fcis-3177 144 5 sets set NOUN fcis-3177 144 6 are be AUX fcis-3177 144 7 divided divide VERB fcis-3177 144 8 into into ADP fcis-3177 144 9 80 80 NUM fcis-3177 144 10 : : SYM fcis-3177 144 11 20 20 NUM fcis-3177 144 12 categories category NOUN fcis-3177 144 13 , , PUNCT fcis-3177 144 14 and and CCONJ fcis-3177 144 15 although although SCONJ fcis-3177 144 16 the the DET fcis-3177 144 17 dataset dataset NOUN fcis-3177 144 18 is be AUX fcis-3177 144 19 more more ADV fcis-3177 144 20 complex complex ADJ fcis-3177 144 21 , , PUNCT fcis-3177 144 22 it it PRON fcis-3177 144 23 is be AUX fcis-3177 144 24 highly highly ADV fcis-3177 144 25 suitable suitable ADJ fcis-3177 144 26 for for ADP fcis-3177 144 27 prototyping prototyping NOUN fcis-3177 144 28 and and CCONJ fcis-3177 144 29 experimental experimental ADJ fcis-3177 144 30 research research NOUN fcis-3177 144 31 . . PUNCT fcis-3177 145 1 2 2 NUM fcis-3177 145 2 ) ) PUNCT fcis-3177 145 3 . . PUNCT fcis-3177 146 1 the the DET fcis-3177 146 2 omniglot omniglot NOUN fcis-3177 146 3 dataset dataset VERB fcis-3177 146 4 [ [ X fcis-3177 146 5 28 28 NUM fcis-3177 146 6 ] ] PUNCT fcis-3177 146 7 , , PUNCT fcis-3177 146 8 which which PRON fcis-3177 146 9 is be AUX fcis-3177 146 10 mainly mainly ADV fcis-3177 146 11 a a DET fcis-3177 146 12 handwritten handwritten ADJ fcis-3177 146 13 dataset dataset NOUN fcis-3177 146 14 consisting consist VERB fcis-3177 146 15 of of ADP fcis-3177 146 16 various various ADJ fcis-3177 146 17 alphabets alphabet NOUN fcis-3177 146 18 , , PUNCT fcis-3177 146 19 was be AUX fcis-3177 146 20 collected collect VERB fcis-3177 146 21 by by ADP fcis-3177 146 22 amazon amazon NOUN fcis-3177 146 23 's 's PART fcis-3177 146 24 mechanical mechanical ADJ fcis-3177 146 25 turk turk NOUN fcis-3177 146 26 . . PUNCT fcis-3177 147 1 unlike unlike ADP fcis-3177 147 2 the the DET fcis-3177 147 3 mnist mnist NOUN fcis-3177 147 4 dataset dataset NOUN fcis-3177 147 5 published publish VERB fcis-3177 147 6 by by ADP fcis-3177 147 7 lecun lecun PROPN fcis-3177 147 8 , , PUNCT fcis-3177 147 9 this this DET fcis-3177 147 10 dataset dataset NOUN fcis-3177 147 11 has have VERB fcis-3177 147 12 many many ADJ fcis-3177 147 13 categories category NOUN fcis-3177 147 14 but but CCONJ fcis-3177 147 15 contains contain VERB fcis-3177 147 16 fewer few ADJ fcis-3177 147 17 data datum NOUN fcis-3177 147 18 per per ADP fcis-3177 147 19 category category NOUN fcis-3177 147 20 . . PUNCT fcis-3177 148 1 3 3 NUM fcis-3177 148 2 ) ) PUNCT fcis-3177 148 3 . . PUNCT fcis-3177 149 1 the the DET fcis-3177 149 2 tiered tiere VERB fcis-3177 149 3 - - PUNCT fcis-3177 149 4 imagenet imagenet NOUN fcis-3177 149 5 dataset dataset NOUN fcis-3177 149 6 is be AUX fcis-3177 149 7 also also ADV fcis-3177 149 8 extracted extract VERB fcis-3177 149 9 from from ADP fcis-3177 149 10 imagenet imagenet NOUN fcis-3177 149 11 . . PUNCT fcis-3177 150 1 compared compare VERB fcis-3177 150 2 with with ADP fcis-3177 150 3 the the DET fcis-3177 150 4 mini mini ADJ fcis-3177 150 5 - - ADJ fcis-3177 150 6 imagenet imagenet ADJ fcis-3177 150 7 dataset dataset NOUN fcis-3177 150 8 , , PUNCT fcis-3177 150 9 the the DET fcis-3177 150 10 tiered tiere VERB fcis-3177 150 11 - - PUNCT fcis-3177 150 12 imagenet imagenet NOUN fcis-3177 150 13 dataset dataset NOUN fcis-3177 150 14 has have VERB fcis-3177 150 15 more more ADJ fcis-3177 150 16 categories category NOUN fcis-3177 150 17 , , PUNCT fcis-3177 150 18 with with ADP fcis-3177 150 19 608 608 NUM fcis-3177 150 20 categories category NOUN fcis-3177 150 21 . . PUNCT fcis-3177 151 1 4 4 NUM fcis-3177 151 2 ) ) PUNCT fcis-3177 151 3 . . PUNCT fcis-3177 152 1 the the DET fcis-3177 152 2 cifar cifar PROPN fcis-3177 152 3 - - PUNCT fcis-3177 152 4 fs fs NOUN fcis-3177 152 5 dataset dataset NOUN fcis-3177 152 6 , , PUNCT fcis-3177 152 7 known know VERB fcis-3177 152 8 as as SCONJ fcis-3177 152 9 the the DET fcis-3177 152 10 cifar100 cifar100 PROPN fcis-3177 152 11 fewshots fewshot NOUN fcis-3177 152 12 dataset dataset VERB fcis-3177 152 13 , , PUNCT fcis-3177 152 14 is be AUX fcis-3177 152 15 extracted extract VERB fcis-3177 152 16 from from ADP fcis-3177 152 17 the the DET fcis-3177 152 18 cifar100 cifar100 PROPN fcis-3177 152 19 dataset dataset NOUN fcis-3177 152 20 and and CCONJ fcis-3177 152 21 contains contain VERB fcis-3177 152 22 100 100 NUM fcis-3177 152 23 categories category NOUN fcis-3177 152 24 . . PUNCT fcis-3177 153 1 5 5 NUM fcis-3177 153 2 ) ) PUNCT fcis-3177 153 3 . . PUNCT fcis-3177 154 1 the the DET fcis-3177 154 2 cub-200 cub-200 PROPN fcis-3177 154 3 dataset dataset NOUN fcis-3177 154 4 , , PUNCT fcis-3177 154 5 known know VERB fcis-3177 154 6 as as ADP fcis-3177 154 7 the the DET fcis-3177 154 8 caltech caltech PROPN fcis-3177 154 9 - - PUNCT fcis-3177 154 10 ucsd ucsd PROPN fcis-3177 154 11 birds-200 birds-200 PROPN fcis-3177 154 12 - - PUNCT fcis-3177 154 13 2011 2011 NUM fcis-3177 154 14 dataset dataset NOUN fcis-3177 154 15 , , PUNCT fcis-3177 154 16 is be AUX fcis-3177 154 17 a a DET fcis-3177 154 18 fine fine ADV fcis-3177 154 19 - - PUNCT fcis-3177 154 20 grained grain VERB fcis-3177 154 21 dataset dataset NOUN fcis-3177 154 22 proposed propose VERB fcis-3177 154 23 by by ADP fcis-3177 154 24 caltech caltech NOUN fcis-3177 154 25 in in ADP fcis-3177 154 26 2010 2010 NUM fcis-3177 154 27 and and CCONJ fcis-3177 154 28 is be AUX fcis-3177 154 29 currently currently ADV fcis-3177 154 30 the the DET fcis-3177 154 31 benchmark benchmark ADJ fcis-3177 154 32 image image NOUN fcis-3177 154 33 dataset dataset VERB fcis-3177 154 34 for for ADP fcis-3177 154 35 fine fine ADV fcis-3177 154 36 - - PUNCT fcis-3177 154 37 grained grain VERB fcis-3177 154 38 classification classification NOUN fcis-3177 154 39 and and CCONJ fcis-3177 154 40 recognition recognition NOUN fcis-3177 154 41 research research NOUN fcis-3177 154 42 , , PUNCT fcis-3177 154 43 containing contain VERB fcis-3177 154 44 images image NOUN fcis-3177 154 45 of of ADP fcis-3177 154 46 200 200 NUM fcis-3177 154 47 bird bird NOUN fcis-3177 154 48 species specie NOUN fcis-3177 154 49 . . PUNCT fcis-3177 155 1 6 6 NUM fcis-3177 155 2 ) ) PUNCT fcis-3177 155 3 . . PUNCT fcis-3177 156 1 the the DET fcis-3177 156 2 stanford stanford PROPN fcis-3177 156 3 dogs dog VERB fcis-3177 156 4 dataset dataset VERB fcis-3177 156 5 [ [ PUNCT fcis-3177 156 6 29 29 NUM fcis-3177 156 7 ] ] PUNCT fcis-3177 156 8 contains contain VERB fcis-3177 156 9 20,580 20,580 NUM fcis-3177 156 10 images image NOUN fcis-3177 156 11 of of ADP fcis-3177 156 12 120 120 NUM fcis-3177 156 13 dog dog NOUN fcis-3177 156 14 species specie NOUN fcis-3177 156 15 worldwide worldwide ADV fcis-3177 156 16 and and CCONJ fcis-3177 156 17 is be AUX fcis-3177 156 18 generally generally ADV fcis-3177 156 19 used use VERB fcis-3177 156 20 for for ADP fcis-3177 156 21 finegrained finegraine VERB fcis-3177 156 22 image image NOUN fcis-3177 156 23 classification classification NOUN fcis-3177 156 24 tasks task NOUN fcis-3177 156 25 . . PUNCT fcis-3177 157 1 7 7 NUM fcis-3177 157 2 ) ) PUNCT fcis-3177 157 3 . . PUNCT fcis-3177 158 1 the the DET fcis-3177 158 2 stanford stanford PROPN fcis-3177 158 3 cars car NOUN fcis-3177 158 4 dataset dataset VERB fcis-3177 158 5 of of ADP fcis-3177 158 6 16,185 16,185 NUM fcis-3177 158 7 images image NOUN fcis-3177 158 8 , , PUNCT fcis-3177 158 9 with with ADP fcis-3177 158 10 categories category NOUN fcis-3177 158 11 classified classify VERB fcis-3177 158 12 mainly mainly ADV fcis-3177 158 13 based base VERB fcis-3177 158 14 on on ADP fcis-3177 158 15 the the DET fcis-3177 158 16 car car NOUN fcis-3177 158 17 's 's PART fcis-3177 158 18 make make NOUN fcis-3177 158 19 , , PUNCT fcis-3177 158 20 model model NOUN fcis-3177 158 21 , , PUNCT fcis-3177 158 22 and and CCONJ fcis-3177 158 23 year year NOUN fcis-3177 158 24 , , PUNCT fcis-3177 158 25 includes include VERB fcis-3177 158 26 a a DET fcis-3177 158 27 sample sample NOUN fcis-3177 158 28 of of ADP fcis-3177 158 29 196 196 NUM fcis-3177 158 30 types type NOUN fcis-3177 158 31 of of ADP fcis-3177 158 32 cars car NOUN fcis-3177 158 33 and and CCONJ fcis-3177 158 34 is be AUX fcis-3177 158 35 generally generally ADV fcis-3177 158 36 used use VERB fcis-3177 158 37 for for ADP fcis-3177 158 38 fine fine ADV fcis-3177 158 39 - - PUNCT fcis-3177 158 40 grained grain VERB fcis-3177 158 41 image image NOUN fcis-3177 158 42 classification classification NOUN fcis-3177 158 43 tasks task NOUN fcis-3177 158 44 . . PUNCT fcis-3177 159 1 table table NOUN fcis-3177 159 2 1 1 NUM fcis-3177 159 3 . . PUNCT fcis-3177 159 4 information information NOUN fcis-3177 159 5 on on ADP fcis-3177 159 6 few few ADJ fcis-3177 159 7 - - PUNCT fcis-3177 159 8 shot shot NOUN fcis-3177 159 9 datasets dataset NOUN fcis-3177 159 10 in in ADP fcis-3177 159 11 computer computer NOUN fcis-3177 159 12 vision vision NOUN fcis-3177 159 13 dataset dataset VERB fcis-3177 159 14 image image NOUN fcis-3177 159 15 size size NOUN fcis-3177 159 16 quantity quantity NOUN fcis-3177 159 17 category category NOUN fcis-3177 159 18 mini mini NOUN fcis-3177 159 19 - - NOUN fcis-3177 159 20 imagenet imagenet ADJ fcis-3177 159 21 84Γ—84 84Γ—84 NUM fcis-3177 159 22 60,000 60,000 NUM fcis-3177 159 23 100 100 NUM fcis-3177 159 24 omniglot omniglot NOUN fcis-3177 159 25 28Γ—28 28Γ—28 NUM fcis-3177 159 26 32,460 32,460 NUM fcis-3177 159 27 1623 1623 NUM fcis-3177 159 28 tiered tiere VERB fcis-3177 159 29 - - PUNCT fcis-3177 159 30 imagenet imagenet NOUN fcis-3177 159 31 84Γ—84 84Γ—84 ADV fcis-3177 159 32 778,848 778,848 NUM fcis-3177 159 33 608 608 NUM fcis-3177 159 34 cifar cifar ADV fcis-3177 159 35 - - PUNCT fcis-3177 159 36 fs fs NOUN fcis-3177 159 37 32Γ—32 32Γ—32 NUM fcis-3177 159 38 60,000 60,000 NUM fcis-3177 159 39 100 100 NUM fcis-3177 159 40 cub-200 cub-200 VERB fcis-3177 159 41 84Γ—84 84Γ—84 NUM fcis-3177 159 42 11,788 11,788 NUM fcis-3177 159 43 200 200 NUM fcis-3177 159 44 stanford stanford PROPN fcis-3177 159 45 dogs dog NOUN fcis-3177 159 46 no no DET fcis-3177 159 47 fixed fix VERB fcis-3177 159 48 size size NOUN fcis-3177 159 49 20,580 20,580 NUM fcis-3177 159 50 120 120 NUM fcis-3177 159 51 stanford stanford PROPN fcis-3177 159 52 cars car NOUN fcis-3177 159 53 360Γ—240 360Γ—240 NUM fcis-3177 159 54 16,185 16,185 NUM fcis-3177 159 55 196 196 NUM fcis-3177 159 56 4.2 4.2 NUM fcis-3177 159 57 . . PUNCT fcis-3177 159 58 natural natural ADJ fcis-3177 159 59 language language NOUN fcis-3177 159 60 processing processing NOUN fcis-3177 159 61 field field NOUN fcis-3177 159 62 1 1 NUM fcis-3177 159 63 ) ) PUNCT fcis-3177 159 64 . . PUNCT fcis-3177 160 1 the the DET fcis-3177 160 2 fewrel fewrel NOUN fcis-3177 160 3 dataset dataset VERB fcis-3177 160 4 [ [ X fcis-3177 160 5 30 30 NUM fcis-3177 160 6 ] ] PUNCT fcis-3177 160 7 proposed propose VERB fcis-3177 160 8 by by ADP fcis-3177 160 9 han han PROPN fcis-3177 160 10 et et PROPN fcis-3177 160 11 al al PROPN fcis-3177 160 12 . . PROPN fcis-3177 161 1 in in ADP fcis-3177 161 2 2018 2018 NUM fcis-3177 161 3 is be AUX fcis-3177 161 4 a a DET fcis-3177 161 5 few few ADJ fcis-3177 161 6 - - PUNCT fcis-3177 161 7 sample sample NOUN fcis-3177 161 8 relationship relationship NOUN fcis-3177 161 9 classification classification NOUN fcis-3177 161 10 dataset dataset NOUN fcis-3177 161 11 containing contain VERB fcis-3177 161 12 64 64 NUM fcis-3177 161 13 relationships relationship NOUN fcis-3177 161 14 for for ADP fcis-3177 161 15 training training NOUN fcis-3177 161 16 , , PUNCT fcis-3177 161 17 16 16 NUM fcis-3177 161 18 relationships relationship NOUN fcis-3177 161 19 for for ADP fcis-3177 161 20 validation validation NOUN fcis-3177 161 21 , , PUNCT fcis-3177 161 22 and and CCONJ fcis-3177 161 23 20 20 NUM fcis-3177 161 24 relationships relationship NOUN fcis-3177 161 25 for for ADP fcis-3177 161 26 testing testing NOUN fcis-3177 161 27 , , PUNCT fcis-3177 161 28 with with ADP fcis-3177 161 29 700 700 NUM fcis-3177 161 30 samples sample NOUN fcis-3177 161 31 under under ADP fcis-3177 161 32 each each DET fcis-3177 161 33 relationship relationship NOUN fcis-3177 161 34 . . PUNCT fcis-3177 162 1 2 2 NUM fcis-3177 162 2 ) ) PUNCT fcis-3177 162 3 . . PUNCT fcis-3177 163 1 the the DET fcis-3177 163 2 arsc arsc ADJ fcis-3177 163 3 dataset dataset NOUN fcis-3177 164 1 [ [ X fcis-3177 164 2 31 31 NUM fcis-3177 164 3 ] ] PUNCT fcis-3177 164 4 proposed propose VERB fcis-3177 164 5 by by ADP fcis-3177 164 6 yu yu PROPN fcis-3177 164 7 et et PROPN fcis-3177 164 8 al al PROPN fcis-3177 164 9 . . PROPN fcis-3177 165 1 in in ADP fcis-3177 165 2 2018 2018 NUM fcis-3177 165 3 is be AUX fcis-3177 165 4 taken take VERB fcis-3177 165 5 from from ADP fcis-3177 165 6 amazon amazon PROPN fcis-3177 165 7 multi multi ADJ fcis-3177 165 8 - - ADJ fcis-3177 165 9 domain domain ADJ fcis-3177 165 10 sentiment sentiment NOUN fcis-3177 165 11 classification classification NOUN fcis-3177 165 12 data datum NOUN fcis-3177 165 13 , , PUNCT fcis-3177 165 14 which which PRON fcis-3177 165 15 contains contain VERB fcis-3177 165 16 review review NOUN fcis-3177 165 17 data datum NOUN fcis-3177 165 18 of of ADP fcis-3177 165 19 23 23 NUM fcis-3177 165 20 amazon amazon NOUN fcis-3177 165 21 items item NOUN fcis-3177 165 22 . . PUNCT fcis-3177 166 1 for for ADP fcis-3177 166 2 114 114 NUM fcis-3177 166 3 each each DET fcis-3177 166 4 item item NOUN fcis-3177 166 5 , , PUNCT fcis-3177 166 6 three three NUM fcis-3177 166 7 binary binary ADJ fcis-3177 166 8 classification classification NOUN fcis-3177 166 9 tasks task NOUN fcis-3177 166 10 are be AUX fcis-3177 166 11 constructed construct VERB fcis-3177 166 12 . . PUNCT fcis-3177 167 1 the the DET fcis-3177 167 2 detailed detailed ADJ fcis-3177 167 3 approach approach NOUN fcis-3177 167 4 is be AUX fcis-3177 167 5 to to PART fcis-3177 167 6 classify classify VERB fcis-3177 167 7 their their PRON fcis-3177 167 8 reviews review NOUN fcis-3177 167 9 into into ADP fcis-3177 167 10 three three NUM fcis-3177 167 11 grades grade NOUN fcis-3177 167 12 by by ADP fcis-3177 167 13 rating rate VERB fcis-3177 167 14 5 5 NUM fcis-3177 167 15 , , PUNCT fcis-3177 167 16 4 4 NUM fcis-3177 167 17 and and CCONJ fcis-3177 167 18 2 2 NUM fcis-3177 167 19 , , PUNCT fcis-3177 167 20 and and CCONJ fcis-3177 167 21 each each DET fcis-3177 167 22 grade grade NOUN fcis-3177 167 23 is be AUX fcis-3177 167 24 considered consider VERB fcis-3177 167 25 as as ADP fcis-3177 167 26 a a DET fcis-3177 167 27 binary binary ADJ fcis-3177 167 28 classification classification NOUN fcis-3177 167 29 task task NOUN fcis-3177 167 30 , , PUNCT fcis-3177 167 31 then then ADV fcis-3177 167 32 23Γ—3=69 23Γ—3=69 NUM fcis-3177 167 33 tasks task NOUN fcis-3177 167 34 are be AUX fcis-3177 167 35 generated generate VERB fcis-3177 167 36 , , PUNCT fcis-3177 167 37 then then ADV fcis-3177 167 38 12 12 NUM fcis-3177 167 39 of of ADP fcis-3177 167 40 them they PRON fcis-3177 167 41 ( ( PUNCT fcis-3177 167 42 4Γ—3 4Γ—3 NOUN fcis-3177 167 43 ) ) PUNCT fcis-3177 167 44 are be AUX fcis-3177 167 45 taken take VERB fcis-3177 167 46 as as ADP fcis-3177 167 47 the the DET fcis-3177 167 48 test test NOUN fcis-3177 167 49 set set NOUN fcis-3177 167 50 , , PUNCT fcis-3177 167 51 and and CCONJ fcis-3177 167 52 the the DET fcis-3177 167 53 remaining remain VERB fcis-3177 167 54 57 57 NUM fcis-3177 167 55 tasks task NOUN fcis-3177 167 56 are be AUX fcis-3177 167 57 used use VERB fcis-3177 167 58 as as ADP fcis-3177 167 59 the the DET fcis-3177 167 60 training training NOUN fcis-3177 167 61 set set NOUN fcis-3177 167 62 . . PUNCT fcis-3177 168 1 3 3 NUM fcis-3177 168 2 ) ) PUNCT fcis-3177 168 3 . . PUNCT fcis-3177 169 1 the the DET fcis-3177 169 2 odic odic ADJ fcis-3177 169 3 dataset dataset NOUN fcis-3177 169 4 [ [ X fcis-3177 169 5 32 32 NUM fcis-3177 169 6 ] ] PUNCT fcis-3177 169 7 came come VERB fcis-3177 169 8 from from ADP fcis-3177 169 9 the the DET fcis-3177 169 10 online online ADJ fcis-3177 169 11 logs log NOUN fcis-3177 169 12 of of ADP fcis-3177 169 13 the the DET fcis-3177 169 14 alibaba alibaba PROPN fcis-3177 169 15 conversational conversational ADJ fcis-3177 169 16 platform platform NOUN fcis-3177 169 17 , , PUNCT fcis-3177 169 18 to to PART fcis-3177 169 19 which which PRON fcis-3177 169 20 users user NOUN fcis-3177 169 21 submit submit VERB fcis-3177 169 22 a a DET fcis-3177 169 23 variety variety NOUN fcis-3177 169 24 of of ADP fcis-3177 169 25 different different ADJ fcis-3177 169 26 conversation conversation NOUN fcis-3177 169 27 tasks task NOUN fcis-3177 169 28 and and CCONJ fcis-3177 169 29 a a DET fcis-3177 169 30 variety variety NOUN fcis-3177 169 31 of of ADP fcis-3177 169 32 different different ADJ fcis-3177 169 33 intentions intention NOUN fcis-3177 169 34 , , PUNCT fcis-3177 169 35 but but CCONJ fcis-3177 169 36 with with ADP fcis-3177 169 37 only only ADV fcis-3177 169 38 a a DET fcis-3177 169 39 tiny tiny ADJ fcis-3177 169 40 amount amount NOUN fcis-3177 169 41 of of ADP fcis-3177 169 42 annotated annotate VERB fcis-3177 169 43 data datum NOUN fcis-3177 169 44 for for ADP fcis-3177 169 45 each each DET fcis-3177 169 46 intention intention NOUN fcis-3177 169 47 , , PUNCT fcis-3177 169 48 which which PRON fcis-3177 169 49 forms form VERB fcis-3177 169 50 a a DET fcis-3177 169 51 typical typical ADJ fcis-3177 169 52 few few ADJ fcis-3177 169 53 - - PUNCT fcis-3177 169 54 shot shot NOUN fcis-3177 169 55 learning learning NOUN fcis-3177 169 56 task task NOUN fcis-3177 169 57 . . PUNCT fcis-3177 170 1 the the DET fcis-3177 170 2 dataset dataset NOUN fcis-3177 170 3 contains contain VERB fcis-3177 170 4 216 216 NUM fcis-3177 170 5 intentions intention NOUN fcis-3177 170 6 , , PUNCT fcis-3177 170 7 of of ADP fcis-3177 170 8 which which PRON fcis-3177 170 9 159 159 NUM fcis-3177 170 10 are be AUX fcis-3177 170 11 used use VERB fcis-3177 170 12 for for ADP fcis-3177 170 13 training training NOUN fcis-3177 170 14 and and CCONJ fcis-3177 170 15 57 57 NUM fcis-3177 170 16 for for ADP fcis-3177 170 17 testing testing NOUN fcis-3177 170 18 . . PUNCT fcis-3177 171 1 5 5 X fcis-3177 171 2 . . X fcis-3177 171 3 conclusion conclusion NOUN fcis-3177 171 4 with with ADP fcis-3177 171 5 the the DET fcis-3177 171 6 influence influence NOUN fcis-3177 171 7 of of ADP fcis-3177 171 8 big big ADJ fcis-3177 171 9 data datum NOUN fcis-3177 171 10 , , PUNCT fcis-3177 171 11 deep deep ADJ fcis-3177 171 12 learning learning NOUN fcis-3177 171 13 has have AUX fcis-3177 171 14 achieved achieve VERB fcis-3177 171 15 remarkable remarkable ADJ fcis-3177 171 16 success success NOUN fcis-3177 171 17 in in ADP fcis-3177 171 18 many many ADJ fcis-3177 171 19 tasks task NOUN fcis-3177 171 20 . . PUNCT fcis-3177 172 1 however however ADV fcis-3177 172 2 , , PUNCT fcis-3177 172 3 labeling label VERB fcis-3177 172 4 large large ADJ fcis-3177 172 5 amounts amount NOUN fcis-3177 172 6 of of ADP fcis-3177 172 7 sample sample NOUN fcis-3177 172 8 data datum NOUN fcis-3177 172 9 in in ADP fcis-3177 172 10 many many ADJ fcis-3177 172 11 real real ADJ fcis-3177 172 12 - - PUNCT fcis-3177 172 13 world world NOUN fcis-3177 172 14 scenarios scenario NOUN fcis-3177 172 15 is be AUX fcis-3177 172 16 often often ADV fcis-3177 172 17 labor labor NOUN fcis-3177 172 18 - - PUNCT fcis-3177 172 19 intensive intensive ADJ fcis-3177 172 20 . . PUNCT fcis-3177 173 1 in in ADP fcis-3177 173 2 many many ADJ fcis-3177 173 3 more more ADJ fcis-3177 173 4 scenarios scenario NOUN fcis-3177 173 5 , , PUNCT fcis-3177 173 6 there there PRON fcis-3177 173 7 are be VERB fcis-3177 173 8 not not PART fcis-3177 173 9 sufficient sufficient ADJ fcis-3177 173 10 samples sample NOUN fcis-3177 173 11 available available ADJ fcis-3177 173 12 for for ADP fcis-3177 173 13 training train VERB fcis-3177 173 14 deep deep ADJ fcis-3177 173 15 neural neural ADJ fcis-3177 173 16 networks network NOUN fcis-3177 173 17 . . PUNCT fcis-3177 174 1 to to PART fcis-3177 174 2 break break VERB fcis-3177 174 3 this this DET fcis-3177 174 4 limitation limitation NOUN fcis-3177 174 5 , , PUNCT fcis-3177 174 6 few few ADJ fcis-3177 174 7 - - PUNCT fcis-3177 174 8 shot shot NOUN fcis-3177 174 9 learning learning NOUN fcis-3177 174 10 is be AUX fcis-3177 174 11 essential essential ADJ fcis-3177 174 12 . . PUNCT fcis-3177 175 1 this this DET fcis-3177 175 2 paper paper NOUN fcis-3177 175 3 briefly briefly NOUN fcis-3177 175 4 describes describe VERB fcis-3177 175 5 two two NUM fcis-3177 175 6 deep deep ADJ fcis-3177 175 7 learning learning NOUN fcis-3177 175 8 models model NOUN fcis-3177 175 9 commonly commonly ADV fcis-3177 175 10 used use VERB fcis-3177 175 11 for for ADP fcis-3177 175 12 few few ADJ fcis-3177 175 13 - - PUNCT fcis-3177 175 14 shot shot NOUN fcis-3177 175 15 learning learning NOUN fcis-3177 175 16 and and CCONJ fcis-3177 175 17 focuses focus VERB fcis-3177 175 18 on on ADP fcis-3177 175 19 four four NUM fcis-3177 175 20 of of ADP fcis-3177 175 21 the the DET fcis-3177 175 22 current current ADJ fcis-3177 175 23 mainstreams mainstream NOUN fcis-3177 175 24 few few ADJ fcis-3177 175 25 - - PUNCT fcis-3177 175 26 shot shot NOUN fcis-3177 175 27 learning learning NOUN fcis-3177 175 28 methods method NOUN fcis-3177 175 29 , , PUNCT fcis-3177 175 30 concluding conclude VERB fcis-3177 175 31 with with ADP fcis-3177 175 32 a a DET fcis-3177 175 33 list list NOUN fcis-3177 175 34 of of ADP fcis-3177 175 35 commonly commonly ADV fcis-3177 175 36 used use VERB fcis-3177 175 37 datasets dataset NOUN fcis-3177 175 38 . . PUNCT fcis-3177 176 1 although although SCONJ fcis-3177 176 2 the the DET fcis-3177 176 3 existing exist VERB fcis-3177 176 4 methods method NOUN fcis-3177 176 5 have have AUX fcis-3177 176 6 achieved achieve VERB fcis-3177 176 7 good good ADJ fcis-3177 176 8 results result NOUN fcis-3177 176 9 , , PUNCT fcis-3177 176 10 there there PRON fcis-3177 176 11 is be VERB fcis-3177 176 12 still still ADV fcis-3177 176 13 massive massive ADJ fcis-3177 176 14 space space NOUN fcis-3177 176 15 for for ADP fcis-3177 176 16 improvement improvement NOUN fcis-3177 176 17 . . PUNCT fcis-3177 177 1 therefore therefore ADV fcis-3177 177 2 , , PUNCT fcis-3177 177 3 in in ADP fcis-3177 177 4 view view NOUN fcis-3177 177 5 of of ADP fcis-3177 177 6 some some DET fcis-3177 177 7 shortcomings shortcoming NOUN fcis-3177 177 8 of of ADP fcis-3177 177 9 existing exist VERB fcis-3177 177 10 techniques technique NOUN fcis-3177 177 11 , , PUNCT fcis-3177 177 12 several several ADJ fcis-3177 177 13 future future ADJ fcis-3177 177 14 directions direction NOUN fcis-3177 177 15 for for ADP fcis-3177 177 16 the the DET fcis-3177 177 17 development development NOUN fcis-3177 177 18 of of ADP fcis-3177 177 19 few few ADJ fcis-3177 177 20 - - PUNCT fcis-3177 177 21 shot shot NOUN fcis-3177 177 22 learning learning NOUN fcis-3177 177 23 are be AUX fcis-3177 177 24 proposed propose VERB fcis-3177 177 25 according accord VERB fcis-3177 177 26 to to ADP fcis-3177 177 27 the the DET fcis-3177 177 28 latest late ADJ fcis-3177 177 29 progress progress NOUN fcis-3177 177 30 in in ADP fcis-3177 177 31 the the DET fcis-3177 177 32 field field NOUN fcis-3177 177 33 of of ADP fcis-3177 177 34 computer computer NOUN fcis-3177 177 35 vision vision NOUN fcis-3177 177 36 : : PUNCT fcis-3177 177 37 ( ( PUNCT fcis-3177 177 38 1 1 X fcis-3177 177 39 ) ) PUNCT fcis-3177 177 40 the the DET fcis-3177 177 41 existing exist VERB fcis-3177 177 42 few few ADJ fcis-3177 177 43 - - PUNCT fcis-3177 177 44 shot shot NOUN fcis-3177 177 45 learning learning NOUN fcis-3177 177 46 models model NOUN fcis-3177 177 47 are be AUX fcis-3177 177 48 based base VERB fcis-3177 177 49 on on ADP fcis-3177 177 50 a a DET fcis-3177 177 51 single single ADJ fcis-3177 177 52 method method NOUN fcis-3177 177 53 . . PUNCT fcis-3177 178 1 in in ADP fcis-3177 178 2 the the DET fcis-3177 178 3 future future NOUN fcis-3177 178 4 , , PUNCT fcis-3177 178 5 we we PRON fcis-3177 178 6 can can AUX fcis-3177 178 7 integrate integrate VERB fcis-3177 178 8 different different ADJ fcis-3177 178 9 fewshot fewshot ADJ fcis-3177 178 10 learning learning NOUN fcis-3177 178 11 methods method NOUN fcis-3177 178 12 , , PUNCT fcis-3177 178 13 focus focus VERB fcis-3177 178 14 on on ADP fcis-3177 178 15 their their PRON fcis-3177 178 16 strengths strength NOUN fcis-3177 178 17 and and CCONJ fcis-3177 178 18 weaknesses weakness NOUN fcis-3177 178 19 , , PUNCT fcis-3177 178 20 and and CCONJ fcis-3177 178 21 improve improve VERB fcis-3177 178 22 to to PART fcis-3177 178 23 achieve achieve VERB fcis-3177 178 24 better well ADJ fcis-3177 178 25 results result NOUN fcis-3177 178 26 . . PUNCT fcis-3177 179 1 ( ( PUNCT fcis-3177 179 2 2 2 X fcis-3177 179 3 ) ) PUNCT fcis-3177 179 4 meta meta NOUN fcis-3177 179 5 - - PUNCT fcis-3177 179 6 learning learning NOUN fcis-3177 179 7 is be AUX fcis-3177 179 8 a a DET fcis-3177 179 9 crucial crucial ADJ fcis-3177 179 10 technique technique NOUN fcis-3177 179 11 to to PART fcis-3177 179 12 solve solve VERB fcis-3177 179 13 the the DET fcis-3177 179 14 problem problem NOUN fcis-3177 179 15 of of ADP fcis-3177 179 16 few few ADJ fcis-3177 179 17 - - PUNCT fcis-3177 179 18 shot shot NOUN fcis-3177 179 19 learning learning NOUN fcis-3177 179 20 , , PUNCT fcis-3177 179 21 but but CCONJ fcis-3177 179 22 the the DET fcis-3177 179 23 current current ADJ fcis-3177 179 24 models model NOUN fcis-3177 179 25 need need VERB fcis-3177 179 26 to to PART fcis-3177 179 27 be be AUX fcis-3177 179 28 more more ADV fcis-3177 179 29 mature mature ADJ fcis-3177 179 30 . . PUNCT fcis-3177 180 1 it it PRON fcis-3177 180 2 is be AUX fcis-3177 180 3 necessary necessary ADJ fcis-3177 180 4 to to PART fcis-3177 180 5 strengthen strengthen VERB fcis-3177 180 6 further far ADV fcis-3177 180 7 the the DET fcis-3177 180 8 research research NOUN fcis-3177 180 9 on on ADP fcis-3177 180 10 meta meta NOUN fcis-3177 180 11 - - PUNCT fcis-3177 180 12 learning learning NOUN fcis-3177 180 13 in in ADP fcis-3177 180 14 the the DET fcis-3177 180 15 field field NOUN fcis-3177 180 16 of of ADP fcis-3177 180 17 few few ADJ fcis-3177 180 18 - - PUNCT fcis-3177 180 19 shot shot NOUN fcis-3177 180 20 learning learning NOUN fcis-3177 180 21 to to PART fcis-3177 180 22 make make VERB fcis-3177 180 23 the the DET fcis-3177 180 24 models model NOUN fcis-3177 180 25 more more ADV fcis-3177 180 26 adaptable adaptable ADJ fcis-3177 180 27 , , PUNCT fcis-3177 180 28 reduce reduce VERB fcis-3177 180 29 overfitting overfitting NOUN fcis-3177 180 30 and and CCONJ fcis-3177 180 31 find find VERB fcis-3177 180 32 better well ADJ fcis-3177 180 33 and and CCONJ fcis-3177 180 34 more more ADV fcis-3177 180 35 stable stable ADJ fcis-3177 180 36 parameters parameter NOUN fcis-3177 180 37 . . PUNCT fcis-3177 181 1 references reference NOUN fcis-3177 181 2 [ [ X fcis-3177 181 3 1 1 NUM fcis-3177 181 4 ] ] X fcis-3177 181 5 biederman biederman NOUN fcis-3177 181 6 , , PUNCT fcis-3177 181 7 i. i. NOUN fcis-3177 181 8 ( ( PUNCT fcis-3177 181 9 1987 1987 NUM fcis-3177 181 10 ) ) PUNCT fcis-3177 181 11 . . PUNCT fcis-3177 182 1 recognition recognition NOUN fcis-3177 182 2 - - PUNCT fcis-3177 182 3 by by ADP fcis-3177 182 4 - - PUNCT fcis-3177 182 5 components component NOUN fcis-3177 182 6 : : PUNCT fcis-3177 182 7 a a DET fcis-3177 182 8 theory theory NOUN fcis-3177 182 9 of of ADP fcis-3177 182 10 human human ADJ fcis-3177 182 11 image image NOUN fcis-3177 182 12 understanding understanding NOUN fcis-3177 182 13 . . PUNCT fcis-3177 183 1 psychological psychological ADJ fcis-3177 183 2 review review NOUN fcis-3177 183 3 , , PUNCT fcis-3177 183 4 94(2 94(2 NOUN fcis-3177 183 5 ) ) PUNCT fcis-3177 183 6 , , PUNCT fcis-3177 183 7 115–147 115–147 NUM fcis-3177 183 8 . . PUNCT fcis-3177 184 1 https://doi.org/10.1037/0033-295x.94.2.115 https://doi.org/10.1037/0033-295x.94.2.115 PROPN fcis-3177 184 2 [ [ X fcis-3177 184 3 2 2 NUM fcis-3177 184 4 ] ] X fcis-3177 184 5 fu fu NOUN fcis-3177 184 6 , , PUNCT fcis-3177 184 7 y. y. PROPN fcis-3177 184 8 , , PUNCT fcis-3177 184 9 xiang xiang PROPN fcis-3177 184 10 , , PUNCT fcis-3177 184 11 l. l. PROPN fcis-3177 184 12 , , PUNCT fcis-3177 184 13 zahid zahid PROPN fcis-3177 184 14 , , PUNCT fcis-3177 184 15 y. y. PROPN fcis-3177 184 16 , , PUNCT fcis-3177 184 17 ding ding NOUN fcis-3177 184 18 , , PUNCT fcis-3177 184 19 g. g. PROPN fcis-3177 184 20 , , PUNCT fcis-3177 184 21 mei mei PROPN fcis-3177 184 22 , , PUNCT fcis-3177 184 23 t. t. PROPN fcis-3177 184 24 , , PUNCT fcis-3177 184 25 shen shen PROPN fcis-3177 184 26 , , PUNCT fcis-3177 184 27 q. q. PROPN fcis-3177 184 28 , , PUNCT fcis-3177 184 29 & & CCONJ fcis-3177 184 30 han han PROPN fcis-3177 184 31 , , PUNCT fcis-3177 184 32 j. j. PROPN fcis-3177 184 33 ( ( PUNCT fcis-3177 184 34 2022 2022 NUM fcis-3177 184 35 ) ) PUNCT fcis-3177 184 36 . . PUNCT fcis-3177 185 1 long long ADV fcis-3177 185 2 - - PUNCT fcis-3177 185 3 tailed tail VERB fcis-3177 185 4 visual visual ADJ fcis-3177 185 5 recognition recognition NOUN fcis-3177 185 6 with with ADP fcis-3177 185 7 deep deep ADJ fcis-3177 185 8 models model NOUN fcis-3177 185 9 : : PUNCT fcis-3177 185 10 a a DET fcis-3177 185 11 methodological methodological ADJ fcis-3177 185 12 survey survey NOUN fcis-3177 185 13 and and CCONJ fcis-3177 185 14 evaluation evaluation NOUN fcis-3177 185 15 . . PUNCT fcis-3177 186 1 neurocomputing neurocompute VERB fcis-3177 186 2 . . PUNCT fcis-3177 187 1 [ [ X fcis-3177 187 2 3 3 NUM fcis-3177 187 3 ] ] X fcis-3177 187 4 fe fe PROPN fcis-3177 187 5 - - PROPN fcis-3177 187 6 fei fei PROPN fcis-3177 187 7 , , PUNCT fcis-3177 187 8 l. l. PROPN fcis-3177 187 9 ( ( PUNCT fcis-3177 187 10 2003 2003 NUM fcis-3177 187 11 ) ) PUNCT fcis-3177 187 12 . . PUNCT fcis-3177 188 1 a a DET fcis-3177 188 2 bayesian bayesian NOUN fcis-3177 188 3 approach approach NOUN fcis-3177 188 4 to to ADP fcis-3177 188 5 unsupervised unsupervised ADJ fcis-3177 188 6 oneshot oneshot NOUN fcis-3177 188 7 learning learning NOUN fcis-3177 188 8 of of ADP fcis-3177 188 9 object object NOUN fcis-3177 188 10 categories category NOUN fcis-3177 188 11 . . PUNCT fcis-3177 189 1 in in ADP fcis-3177 189 2 proceedings proceeding NOUN fcis-3177 189 3 ninth ninth ADJ fcis-3177 189 4 ieee ieee PROPN fcis-3177 189 5 international international PROPN fcis-3177 189 6 conference conference NOUN fcis-3177 189 7 on on ADP fcis-3177 189 8 computer computer NOUN fcis-3177 189 9 vision vision NOUN fcis-3177 189 10 ( ( PUNCT fcis-3177 189 11 pp pp ADJ fcis-3177 189 12 . . PUNCT fcis-3177 190 1 1134 1134 NUM fcis-3177 190 2 - - SYM fcis-3177 190 3 1141 1141 NUM fcis-3177 190 4 ) ) PUNCT fcis-3177 190 5 . . PUNCT fcis-3177 191 1 ieee ieee PROPN fcis-3177 191 2 . . PUNCT fcis-3177 192 1 [ [ X fcis-3177 192 2 4 4 NUM fcis-3177 192 3 ] ] X fcis-3177 192 4 wang wang PROPN fcis-3177 192 5 , , PUNCT fcis-3177 192 6 y. y. PROPN fcis-3177 192 7 , , PUNCT fcis-3177 192 8 yao yao PROPN fcis-3177 192 9 , , PUNCT fcis-3177 192 10 q. q. PROPN fcis-3177 192 11 , , PUNCT fcis-3177 192 12 kwok kwok PROPN fcis-3177 192 13 , , PUNCT fcis-3177 192 14 j. j. PROPN fcis-3177 192 15 t. t. PROPN fcis-3177 192 16 , , PUNCT fcis-3177 192 17 & & CCONJ fcis-3177 192 18 ni ni PROPN fcis-3177 192 19 , , PUNCT fcis-3177 192 20 l. l. PROPN fcis-3177 192 21 m. m. PROPN fcis-3177 192 22 ( ( PUNCT fcis-3177 192 23 2020 2020 NUM fcis-3177 192 24 ) ) PUNCT fcis-3177 192 25 . . PUNCT fcis-3177 193 1 generalizing generalize VERB fcis-3177 193 2 from from ADP fcis-3177 193 3 a a DET fcis-3177 193 4 few few ADJ fcis-3177 193 5 examples example NOUN fcis-3177 193 6 : : PUNCT fcis-3177 193 7 a a DET fcis-3177 193 8 survey survey NOUN fcis-3177 193 9 on on ADP fcis-3177 193 10 few few ADJ fcis-3177 193 11 - - PUNCT fcis-3177 193 12 shot shot NOUN fcis-3177 193 13 learning learning NOUN fcis-3177 193 14 . . PUNCT fcis-3177 194 1 acm acm PROPN fcis-3177 194 2 computing computing NOUN fcis-3177 194 3 surveys survey NOUN fcis-3177 194 4 ( ( PUNCT fcis-3177 194 5 csur csur NOUN fcis-3177 194 6 ) ) PUNCT fcis-3177 194 7 , , PUNCT fcis-3177 194 8 53(3 53(3 NUM fcis-3177 194 9 ) ) PUNCT fcis-3177 194 10 , , PUNCT fcis-3177 194 11 1 1 NUM fcis-3177 194 12 - - SYM fcis-3177 194 13 34 34 NUM fcis-3177 194 14 . . PUNCT fcis-3177 195 1 [ [ X fcis-3177 195 2 5 5 NUM fcis-3177 195 3 ] ] X fcis-3177 195 4 hinton hinton PROPN fcis-3177 195 5 , , PUNCT fcis-3177 195 6 g. g. PROPN fcis-3177 195 7 e. e. PROPN fcis-3177 195 8 , , PUNCT fcis-3177 195 9 & & CCONJ fcis-3177 195 10 salakhutdinov salakhutdinov PROPN fcis-3177 195 11 , , PUNCT fcis-3177 195 12 r. r. PROPN fcis-3177 195 13 r. r. PROPN fcis-3177 195 14 ( ( PUNCT fcis-3177 195 15 2006 2006 NUM fcis-3177 195 16 ) ) PUNCT fcis-3177 195 17 . . PUNCT fcis-3177 196 1 reducing reduce VERB fcis-3177 196 2 the the DET fcis-3177 196 3 dimensionality dimensionality NOUN fcis-3177 196 4 of of ADP fcis-3177 196 5 data datum NOUN fcis-3177 196 6 with with ADP fcis-3177 196 7 neural neural ADJ fcis-3177 196 8 networks network NOUN fcis-3177 196 9 . . PUNCT fcis-3177 197 1 science science NOUN fcis-3177 197 2 , , PUNCT fcis-3177 197 3 313(5786 313(5786 NUM fcis-3177 197 4 ) ) PUNCT fcis-3177 197 5 , , PUNCT fcis-3177 197 6 504 504 NUM fcis-3177 197 7 - - SYM fcis-3177 197 8 507 507 NUM fcis-3177 197 9 . . PUNCT fcis-3177 198 1 [ [ X fcis-3177 198 2 6 6 NUM fcis-3177 198 3 ] ] PUNCT fcis-3177 198 4 he he PRON fcis-3177 198 5 , , PUNCT fcis-3177 198 6 k. k. PROPN fcis-3177 198 7 , , PUNCT fcis-3177 198 8 zhang zhang PROPN fcis-3177 198 9 , , PUNCT fcis-3177 198 10 x. x. PROPN fcis-3177 198 11 , , PUNCT fcis-3177 198 12 ren ren PROPN fcis-3177 198 13 , , PUNCT fcis-3177 198 14 s. s. PROPN fcis-3177 198 15 , , PUNCT fcis-3177 198 16 & & CCONJ fcis-3177 198 17 sun sun PROPN fcis-3177 198 18 , , PUNCT fcis-3177 198 19 j. j. PROPN fcis-3177 198 20 ( ( PUNCT fcis-3177 198 21 2016 2016 NUM fcis-3177 198 22 ) ) PUNCT fcis-3177 198 23 . . PUNCT fcis-3177 199 1 deep deep ADJ fcis-3177 199 2 residual residual ADJ fcis-3177 199 3 learning learning NOUN fcis-3177 199 4 for for ADP fcis-3177 199 5 image image NOUN fcis-3177 199 6 recognition recognition NOUN fcis-3177 199 7 . . PUNCT fcis-3177 200 1 in in ADP fcis-3177 200 2 proceedings proceeding NOUN fcis-3177 200 3 of of ADP fcis-3177 200 4 the the DET fcis-3177 200 5 ieee ieee NOUN fcis-3177 200 6 conference conference NOUN fcis-3177 200 7 on on ADP fcis-3177 200 8 computer computer NOUN fcis-3177 200 9 vision vision NOUN fcis-3177 200 10 and and CCONJ fcis-3177 200 11 pattern pattern NOUN fcis-3177 200 12 recognition recognition NOUN fcis-3177 200 13 ( ( PUNCT fcis-3177 200 14 pp pp ADJ fcis-3177 200 15 . . PUNCT fcis-3177 201 1 770 770 NUM fcis-3177 201 2 - - SYM fcis-3177 201 3 778 778 NUM fcis-3177 201 4 ) ) PUNCT fcis-3177 201 5 . . PUNCT fcis-3177 202 1 [ [ X fcis-3177 202 2 7 7 X fcis-3177 202 3 ] ] X fcis-3177 202 4 szegedy szegedy PROPN fcis-3177 202 5 , , PUNCT fcis-3177 202 6 c. c. PROPN fcis-3177 202 7 , , PUNCT fcis-3177 202 8 liu liu PROPN fcis-3177 202 9 , , PUNCT fcis-3177 202 10 w. w. PROPN fcis-3177 202 11 , , PUNCT fcis-3177 202 12 jia jia PROPN fcis-3177 202 13 , , PUNCT fcis-3177 202 14 y. y. PROPN fcis-3177 202 15 , , PUNCT fcis-3177 202 16 sermanet sermanet NOUN fcis-3177 202 17 , , PUNCT fcis-3177 202 18 p. p. PROPN fcis-3177 202 19 , , PUNCT fcis-3177 202 20 reed reed PROPN fcis-3177 202 21 , , PUNCT fcis-3177 202 22 s. s. PROPN fcis-3177 202 23 , , PUNCT fcis-3177 202 24 anguelov anguelov PROPN fcis-3177 202 25 , , PUNCT fcis-3177 202 26 d. d. PROPN fcis-3177 202 27 , , PUNCT fcis-3177 202 28 ... ... PUNCT fcis-3177 202 29 & & CCONJ fcis-3177 202 30 rabinovich rabinovich PROPN fcis-3177 202 31 , , PUNCT fcis-3177 202 32 a. a. PROPN fcis-3177 202 33 ( ( PUNCT fcis-3177 202 34 2015 2015 NUM fcis-3177 202 35 ) ) PUNCT fcis-3177 202 36 . . PUNCT fcis-3177 203 1 going go VERB fcis-3177 203 2 deeper deeply ADV fcis-3177 203 3 with with ADP fcis-3177 203 4 convolutions convolution NOUN fcis-3177 203 5 . . PUNCT fcis-3177 204 1 in in ADP fcis-3177 204 2 proceedings proceeding NOUN fcis-3177 204 3 of of ADP fcis-3177 204 4 the the DET fcis-3177 204 5 ieee ieee NOUN fcis-3177 204 6 conference conference NOUN fcis-3177 204 7 on on ADP fcis-3177 204 8 computer computer NOUN fcis-3177 204 9 vision vision NOUN fcis-3177 204 10 and and CCONJ fcis-3177 204 11 pattern pattern NOUN fcis-3177 204 12 recognition recognition NOUN fcis-3177 204 13 ( ( PUNCT fcis-3177 204 14 pp pp ADJ fcis-3177 204 15 . . PUNCT fcis-3177 205 1 1 1 NUM fcis-3177 205 2 - - SYM fcis-3177 205 3 9 9 NUM fcis-3177 205 4 ) ) PUNCT fcis-3177 205 5 . . PUNCT fcis-3177 206 1 [ [ X fcis-3177 206 2 8 8 NUM fcis-3177 206 3 ] ] SYM fcis-3177 206 4 simonyan simonyan ADJ fcis-3177 206 5 , , PUNCT fcis-3177 206 6 k. k. PROPN fcis-3177 206 7 , , PUNCT fcis-3177 206 8 & & CCONJ fcis-3177 206 9 zisserman zisserman PROPN fcis-3177 206 10 , , PUNCT fcis-3177 206 11 a. a. NOUN fcis-3177 206 12 ( ( PUNCT fcis-3177 206 13 2014 2014 NUM fcis-3177 206 14 ) ) PUNCT fcis-3177 206 15 . . PUNCT fcis-3177 207 1 very very ADV fcis-3177 207 2 deep deep ADJ fcis-3177 207 3 convolutional convolutional ADJ fcis-3177 207 4 networks network NOUN fcis-3177 207 5 for for ADP fcis-3177 207 6 large large ADJ fcis-3177 207 7 - - PUNCT fcis-3177 207 8 scale scale NOUN fcis-3177 207 9 image image NOUN fcis-3177 207 10 recognition recognition NOUN fcis-3177 207 11 . . PUNCT fcis-3177 208 1 arxiv arxiv PROPN fcis-3177 208 2 preprint preprint PROPN fcis-3177 208 3 arxiv:1409.1556 arxiv:1409.1556 NOUN fcis-3177 208 4 . . PUNCT fcis-3177 209 1 [ [ X fcis-3177 209 2 9 9 NUM fcis-3177 209 3 ] ] SYM fcis-3177 209 4 lecun lecun ADJ fcis-3177 209 5 , , PUNCT fcis-3177 209 6 y. y. PROPN fcis-3177 209 7 , , PUNCT fcis-3177 209 8 bottou bottou PROPN fcis-3177 209 9 , , PUNCT fcis-3177 209 10 l. l. PROPN fcis-3177 209 11 , , PUNCT fcis-3177 209 12 bengio bengio PROPN fcis-3177 209 13 , , PUNCT fcis-3177 209 14 y. y. PROPN fcis-3177 209 15 , , PUNCT fcis-3177 209 16 & & CCONJ fcis-3177 209 17 haffner haffner PROPN fcis-3177 209 18 , , PUNCT fcis-3177 209 19 p. p. NOUN fcis-3177 209 20 ( ( PUNCT fcis-3177 209 21 1998 1998 NUM fcis-3177 209 22 ) ) PUNCT fcis-3177 209 23 . . PUNCT fcis-3177 210 1 gradient gradient NOUN fcis-3177 210 2 - - PUNCT fcis-3177 210 3 based base VERB fcis-3177 210 4 learning learning NOUN fcis-3177 210 5 applied apply VERB fcis-3177 210 6 to to ADP fcis-3177 210 7 document document NOUN fcis-3177 210 8 recognition recognition NOUN fcis-3177 210 9 . . PUNCT fcis-3177 211 1 proceedings proceeding NOUN fcis-3177 211 2 of of ADP fcis-3177 211 3 the the DET fcis-3177 211 4 ieee ieee NOUN fcis-3177 211 5 , , PUNCT fcis-3177 211 6 86(11 86(11 NUM fcis-3177 211 7 ) ) PUNCT fcis-3177 211 8 , , PUNCT fcis-3177 211 9 2278 2278 NUM fcis-3177 211 10 - - SYM fcis-3177 211 11 2324 2324 NUM fcis-3177 211 12 . . PUNCT fcis-3177 212 1 [ [ X fcis-3177 212 2 10 10 NUM fcis-3177 212 3 ] ] X fcis-3177 212 4 hopfield hopfield PROPN fcis-3177 212 5 , , PUNCT fcis-3177 212 6 j. j. PROPN fcis-3177 212 7 j. j. PROPN fcis-3177 212 8 ( ( PUNCT fcis-3177 212 9 1982 1982 NUM fcis-3177 212 10 ) ) PUNCT fcis-3177 212 11 . . PUNCT fcis-3177 213 1 neural neural ADJ fcis-3177 213 2 networks network NOUN fcis-3177 213 3 and and CCONJ fcis-3177 213 4 physical physical ADJ fcis-3177 213 5 systems system NOUN fcis-3177 213 6 with with ADP fcis-3177 213 7 emergent emergent ADJ fcis-3177 213 8 collective collective ADJ fcis-3177 213 9 computational computational ADJ fcis-3177 213 10 abilities ability NOUN fcis-3177 213 11 . . PUNCT fcis-3177 214 1 proceedings proceeding NOUN fcis-3177 214 2 of of ADP fcis-3177 214 3 the the DET fcis-3177 214 4 national national PROPN fcis-3177 214 5 academy academy PROPN fcis-3177 214 6 of of ADP fcis-3177 214 7 sciences sciences PROPN fcis-3177 214 8 , , PUNCT fcis-3177 214 9 79(8 79(8 NUM fcis-3177 214 10 ) ) PUNCT fcis-3177 214 11 , , PUNCT fcis-3177 214 12 2554 2554 NUM fcis-3177 214 13 - - SYM fcis-3177 214 14 2558 2558 NUM fcis-3177 214 15 . . PUNCT fcis-3177 215 1 [ [ X fcis-3177 215 2 11 11 NUM fcis-3177 215 3 ] ] X fcis-3177 215 4 jordan jordan PROPN fcis-3177 215 5 , , PUNCT fcis-3177 215 6 m. m. NOUN fcis-3177 215 7 i. i. PROPN fcis-3177 215 8 ( ( PUNCT fcis-3177 215 9 1986 1986 NUM fcis-3177 215 10 ) ) PUNCT fcis-3177 215 11 . . PUNCT fcis-3177 216 1 serial serial ADJ fcis-3177 216 2 order order NOUN fcis-3177 216 3 : : PUNCT fcis-3177 216 4 a a DET fcis-3177 216 5 parallel parallel ADJ fcis-3177 216 6 distrmuted distrmute VERB fcis-3177 216 7 processing processing NOUN fcis-3177 216 8 approach approach NOUN fcis-3177 216 9 . . PUNCT fcis-3177 217 1 [ [ X fcis-3177 217 2 12 12 NUM fcis-3177 217 3 ] ] X fcis-3177 217 4 elman elman NOUN fcis-3177 217 5 , , PUNCT fcis-3177 217 6 j. j. PROPN fcis-3177 217 7 l. l. PROPN fcis-3177 217 8 ( ( PUNCT fcis-3177 217 9 1990 1990 NUM fcis-3177 217 10 ) ) PUNCT fcis-3177 217 11 . . PUNCT fcis-3177 218 1 finding find VERB fcis-3177 218 2 structure structure NOUN fcis-3177 218 3 in in ADP fcis-3177 218 4 time time NOUN fcis-3177 218 5 . . PUNCT fcis-3177 219 1 cognitive cognitive ADJ fcis-3177 219 2 science science NOUN fcis-3177 219 3 , , PUNCT fcis-3177 219 4 14(2 14(2 NUM fcis-3177 219 5 ) ) PUNCT fcis-3177 219 6 , , PUNCT fcis-3177 219 7 179 179 NUM fcis-3177 219 8 - - SYM fcis-3177 219 9 211 211 NUM fcis-3177 219 10 . . PUNCT fcis-3177 220 1 [ [ X fcis-3177 220 2 13 13 NUM fcis-3177 220 3 ] ] SYM fcis-3177 220 4 palangi palangi PROPN fcis-3177 220 5 , , PUNCT fcis-3177 220 6 h. h. PROPN fcis-3177 220 7 , , PUNCT fcis-3177 220 8 deng deng PROPN fcis-3177 220 9 , , PUNCT fcis-3177 220 10 l. l. PROPN fcis-3177 220 11 , , PUNCT fcis-3177 220 12 shen shen PROPN fcis-3177 220 13 , , PUNCT fcis-3177 220 14 y. y. PROPN fcis-3177 220 15 , , PUNCT fcis-3177 220 16 gao gao PROPN fcis-3177 220 17 , , PUNCT fcis-3177 220 18 j. j. PROPN fcis-3177 220 19 , , PUNCT fcis-3177 220 20 he he PRON fcis-3177 220 21 , , PUNCT fcis-3177 220 22 x. x. PROPN fcis-3177 220 23 , , PUNCT fcis-3177 220 24 chen chen PROPN fcis-3177 220 25 , , PUNCT fcis-3177 220 26 j. j. PROPN fcis-3177 220 27 , , PUNCT fcis-3177 220 28 ... ... PUNCT fcis-3177 220 29 & & CCONJ fcis-3177 220 30 ward ward PROPN fcis-3177 220 31 , , PUNCT fcis-3177 220 32 r. r. PROPN fcis-3177 220 33 ( ( PUNCT fcis-3177 220 34 2016 2016 NUM fcis-3177 220 35 ) ) PUNCT fcis-3177 220 36 . . PUNCT fcis-3177 221 1 deep deep ADJ fcis-3177 221 2 sentence sentence NOUN fcis-3177 221 3 embedding embed VERB fcis-3177 221 4 using use VERB fcis-3177 221 5 long long ADJ fcis-3177 221 6 shortterm shortterm PROPN fcis-3177 221 7 memory memory NOUN fcis-3177 221 8 networks network NOUN fcis-3177 221 9 : : PUNCT fcis-3177 221 10 analysis analysis NOUN fcis-3177 221 11 and and CCONJ fcis-3177 221 12 application application NOUN fcis-3177 221 13 to to ADP fcis-3177 221 14 information information NOUN fcis-3177 221 15 retrieval retrieval NOUN fcis-3177 221 16 . . PUNCT fcis-3177 222 1 ieee ieee PROPN fcis-3177 222 2 / / SYM fcis-3177 222 3 acm acm PROPN fcis-3177 222 4 transactions transaction NOUN fcis-3177 222 5 on on ADP fcis-3177 222 6 audio audio NOUN fcis-3177 222 7 , , PUNCT fcis-3177 222 8 speech speech NOUN fcis-3177 222 9 , , PUNCT fcis-3177 222 10 and and CCONJ fcis-3177 222 11 language language NOUN fcis-3177 222 12 processing processing NOUN fcis-3177 222 13 , , PUNCT fcis-3177 222 14 24(4 24(4 NOUN fcis-3177 222 15 ) ) PUNCT fcis-3177 222 16 , , PUNCT fcis-3177 222 17 694 694 NUM fcis-3177 222 18 - - SYM fcis-3177 222 19 707 707 NUM fcis-3177 222 20 . . PUNCT fcis-3177 223 1 [ [ X fcis-3177 223 2 14 14 NUM fcis-3177 223 3 ] ] X fcis-3177 223 4 hariharan hariharan PROPN fcis-3177 223 5 , , PUNCT fcis-3177 223 6 b. b. PROPN fcis-3177 223 7 , , PUNCT fcis-3177 223 8 & & CCONJ fcis-3177 223 9 girshick girshick PROPN fcis-3177 223 10 , , PUNCT fcis-3177 223 11 r. r. PROPN fcis-3177 223 12 ( ( PUNCT fcis-3177 223 13 2017 2017 NUM fcis-3177 223 14 ) ) PUNCT fcis-3177 223 15 . . PUNCT fcis-3177 224 1 low low ADJ fcis-3177 224 2 - - PUNCT fcis-3177 224 3 shot shot NOUN fcis-3177 224 4 visual visual ADJ fcis-3177 224 5 recognition recognition NOUN fcis-3177 224 6 by by ADP fcis-3177 224 7 shrinking shrink VERB fcis-3177 224 8 and and CCONJ fcis-3177 224 9 hallucinating hallucinate VERB fcis-3177 224 10 features feature NOUN fcis-3177 224 11 . . PUNCT fcis-3177 225 1 in in ADP fcis-3177 225 2 proceedings proceeding NOUN fcis-3177 225 3 of of ADP fcis-3177 225 4 the the DET fcis-3177 225 5 ieee ieee NOUN fcis-3177 225 6 international international PROPN fcis-3177 225 7 conference conference NOUN fcis-3177 225 8 on on ADP fcis-3177 225 9 computer computer NOUN fcis-3177 225 10 vision vision NOUN fcis-3177 225 11 ( ( PUNCT fcis-3177 225 12 pp pp ADJ fcis-3177 225 13 . . PUNCT fcis-3177 225 14 3018 3018 NUM fcis-3177 225 15 - - SYM fcis-3177 225 16 3027 3027 NUM fcis-3177 225 17 ) ) PUNCT fcis-3177 225 18 . . PUNCT fcis-3177 226 1 [ [ X fcis-3177 226 2 15 15 NUM fcis-3177 226 3 ] ] X fcis-3177 226 4 wang wang PROPN fcis-3177 226 5 , , PUNCT fcis-3177 226 6 y. y. PROPN fcis-3177 226 7 x. x. PROPN fcis-3177 226 8 , , PUNCT fcis-3177 226 9 girshick girshick PROPN fcis-3177 226 10 , , PUNCT fcis-3177 226 11 r. r. PROPN fcis-3177 226 12 , , PUNCT fcis-3177 226 13 hebert hebert NOUN fcis-3177 226 14 , , PUNCT fcis-3177 226 15 m. m. NOUN fcis-3177 226 16 , , PUNCT fcis-3177 226 17 & & CCONJ fcis-3177 226 18 hariharan hariharan PROPN fcis-3177 226 19 , , PUNCT fcis-3177 226 20 b. b. PROPN fcis-3177 226 21 ( ( PUNCT fcis-3177 226 22 2018 2018 NUM fcis-3177 226 23 ) ) PUNCT fcis-3177 226 24 . . PUNCT fcis-3177 227 1 low low ADJ fcis-3177 227 2 - - PUNCT fcis-3177 227 3 shot shot NOUN fcis-3177 227 4 learning learning NOUN fcis-3177 227 5 from from ADP fcis-3177 227 6 imaginary imaginary ADJ fcis-3177 227 7 data datum NOUN fcis-3177 227 8 . . PUNCT fcis-3177 228 1 in in ADP fcis-3177 228 2 proceedings proceeding NOUN fcis-3177 228 3 of of ADP fcis-3177 228 4 the the DET fcis-3177 228 5 ieee ieee NOUN fcis-3177 228 6 conference conference NOUN fcis-3177 228 7 on on ADP fcis-3177 228 8 computer computer NOUN fcis-3177 228 9 vision vision NOUN fcis-3177 228 10 and and CCONJ fcis-3177 228 11 pattern pattern NOUN fcis-3177 228 12 recognition recognition NOUN fcis-3177 228 13 ( ( PUNCT fcis-3177 228 14 pp pp ADJ fcis-3177 228 15 . . PUNCT fcis-3177 229 1 7278 7278 NUM fcis-3177 229 2 - - SYM fcis-3177 229 3 7286 7286 NUM fcis-3177 229 4 ) ) PUNCT fcis-3177 229 5 . . PUNCT fcis-3177 230 1 [ [ X fcis-3177 230 2 16 16 NUM fcis-3177 230 3 ] ] X fcis-3177 230 4 howard howard PROPN fcis-3177 230 5 , , PUNCT fcis-3177 230 6 j. j. PROPN fcis-3177 230 7 , , PUNCT fcis-3177 230 8 & & CCONJ fcis-3177 230 9 ruder ruder PROPN fcis-3177 230 10 , , PUNCT fcis-3177 230 11 s. s. PROPN fcis-3177 230 12 ( ( PUNCT fcis-3177 230 13 2018 2018 NUM fcis-3177 230 14 ) ) PUNCT fcis-3177 230 15 . . PUNCT fcis-3177 231 1 universal universal ADJ fcis-3177 231 2 language language NOUN fcis-3177 231 3 model model NOUN fcis-3177 231 4 fine fine ADV fcis-3177 231 5 - - PUNCT fcis-3177 231 6 tuning tuning NOUN fcis-3177 231 7 for for ADP fcis-3177 231 8 text text NOUN fcis-3177 231 9 classification classification NOUN fcis-3177 231 10 . . PUNCT fcis-3177 232 1 arxiv arxiv PROPN fcis-3177 232 2 preprint preprint NOUN fcis-3177 232 3 arxiv:1801.06146 arxiv:1801.06146 NOUN fcis-3177 232 4 . . PUNCT fcis-3177 233 1 [ [ X fcis-3177 233 2 17 17 NUM fcis-3177 233 3 ] ] PUNCT fcis-3177 233 4 nakamura nakamura NOUN fcis-3177 233 5 , , PUNCT fcis-3177 233 6 a. a. PROPN fcis-3177 233 7 , , PUNCT fcis-3177 233 8 & & CCONJ fcis-3177 233 9 harada harada PROPN fcis-3177 233 10 , , PUNCT fcis-3177 233 11 t. t. PROPN fcis-3177 233 12 ( ( PUNCT fcis-3177 233 13 2019 2019 NUM fcis-3177 233 14 ) ) PUNCT fcis-3177 233 15 . . PUNCT fcis-3177 234 1 revisiting revisit VERB fcis-3177 234 2 fine fine ADV fcis-3177 234 3 - - PUNCT fcis-3177 234 4 tuning tuning NOUN fcis-3177 234 5 for for ADP fcis-3177 234 6 few few ADJ fcis-3177 234 7 - - PUNCT fcis-3177 234 8 shot shot NOUN fcis-3177 234 9 learning learning NOUN fcis-3177 234 10 . . PUNCT fcis-3177 235 1 arxiv arxiv PROPN fcis-3177 235 2 preprint preprint NOUN fcis-3177 235 3 arxiv:1910.00216 arxiv:1910.00216 VERB fcis-3177 235 4 . . PUNCT fcis-3177 236 1 [ [ X fcis-3177 236 2 18 18 NUM fcis-3177 236 3 ] ] PUNCT fcis-3177 236 4 bellet bellet NOUN fcis-3177 236 5 , , PUNCT fcis-3177 236 6 a. a. NOUN fcis-3177 236 7 , , PUNCT fcis-3177 236 8 habrard habrard PROPN fcis-3177 236 9 , , PUNCT fcis-3177 236 10 a. a. NOUN fcis-3177 236 11 , , PUNCT fcis-3177 236 12 & & CCONJ fcis-3177 236 13 sebban sebban ADJ fcis-3177 236 14 , , PUNCT fcis-3177 236 15 m. m. NOUN fcis-3177 236 16 ( ( PUNCT fcis-3177 236 17 2013 2013 NUM fcis-3177 236 18 ) ) PUNCT fcis-3177 236 19 . . PUNCT fcis-3177 237 1 a a DET fcis-3177 237 2 survey survey NOUN fcis-3177 237 3 on on ADP fcis-3177 237 4 metric metric ADJ fcis-3177 237 5 learning learning NOUN fcis-3177 237 6 for for ADP fcis-3177 237 7 feature feature NOUN fcis-3177 237 8 vectors vector NOUN fcis-3177 237 9 and and CCONJ fcis-3177 237 10 structured structured ADJ fcis-3177 237 11 data datum NOUN fcis-3177 237 12 . . PUNCT fcis-3177 238 1 arxiv arxiv PROPN fcis-3177 238 2 preprint preprint VERB fcis-3177 238 3 arxiv:1306.6709 arxiv:1306.6709 PROPN fcis-3177 238 4 . . PUNCT fcis-3177 239 1 [ [ X fcis-3177 239 2 19 19 NUM fcis-3177 239 3 ] ] X fcis-3177 239 4 weinberger weinberger PROPN fcis-3177 239 5 , , PUNCT fcis-3177 239 6 k. k. PROPN fcis-3177 239 7 q. q. PROPN fcis-3177 239 8 , , PUNCT fcis-3177 239 9 & & CCONJ fcis-3177 239 10 saul saul PROPN fcis-3177 239 11 , , PUNCT fcis-3177 239 12 l. l. PROPN fcis-3177 239 13 k. k. PROPN fcis-3177 239 14 ( ( PUNCT fcis-3177 239 15 2009 2009 NUM fcis-3177 239 16 ) ) PUNCT fcis-3177 239 17 . . PUNCT fcis-3177 240 1 distance distance NOUN fcis-3177 240 2 metric metric ADJ fcis-3177 240 3 learning learning NOUN fcis-3177 240 4 for for ADP fcis-3177 240 5 large large ADJ fcis-3177 240 6 margin margin NOUN fcis-3177 240 7 nearest near ADJ fcis-3177 240 8 neighbor neighbor NOUN fcis-3177 240 9 classification classification NOUN fcis-3177 240 10 . . PUNCT fcis-3177 241 1 journal journal NOUN fcis-3177 241 2 of of ADP fcis-3177 241 3 machine machine NOUN fcis-3177 241 4 learning learn VERB fcis-3177 241 5 research research NOUN fcis-3177 241 6 , , PUNCT fcis-3177 241 7 10(2 10(2 NUM fcis-3177 241 8 ) ) PUNCT fcis-3177 241 9 . . PUNCT fcis-3177 242 1 [ [ X fcis-3177 242 2 20 20 NUM fcis-3177 242 3 ] ] X fcis-3177 242 4 koch koch PROPN fcis-3177 242 5 , , PUNCT fcis-3177 242 6 g. g. PROPN fcis-3177 242 7 , , PUNCT fcis-3177 242 8 zemel zemel PROPN fcis-3177 242 9 , , PUNCT fcis-3177 242 10 r. r. PROPN fcis-3177 242 11 , , PUNCT fcis-3177 242 12 & & CCONJ fcis-3177 242 13 salakhutdinov salakhutdinov PROPN fcis-3177 242 14 , , PUNCT fcis-3177 242 15 r. r. PROPN fcis-3177 242 16 ( ( PUNCT fcis-3177 242 17 2015 2015 NUM fcis-3177 242 18 , , PUNCT fcis-3177 242 19 july july PROPN fcis-3177 242 20 ) ) PUNCT fcis-3177 242 21 . . PUNCT fcis-3177 243 1 siamese siamese ADJ fcis-3177 243 2 neural neural ADJ fcis-3177 243 3 networks network NOUN fcis-3177 243 4 for for ADP fcis-3177 243 5 one one NUM fcis-3177 243 6 - - PUNCT fcis-3177 243 7 shot shot NOUN fcis-3177 243 8 image image NOUN fcis-3177 243 9 recognition recognition NOUN fcis-3177 243 10 . . PUNCT fcis-3177 244 1 in in ADP fcis-3177 244 2 icml icml NOUN fcis-3177 244 3 deep deep ADJ fcis-3177 244 4 learning learning NOUN fcis-3177 244 5 workshop workshop NOUN fcis-3177 244 6 ( ( PUNCT fcis-3177 244 7 vol vol NOUN fcis-3177 244 8 . . NOUN fcis-3177 244 9 2 2 NUM fcis-3177 244 10 , , PUNCT fcis-3177 244 11 p. p. NOUN fcis-3177 244 12 0 0 NUM fcis-3177 244 13 ) ) PUNCT fcis-3177 244 14 . . PUNCT fcis-3177 245 1 [ [ X fcis-3177 245 2 21 21 NUM fcis-3177 245 3 ] ] X fcis-3177 245 4 santoro santoro PROPN fcis-3177 245 5 , , PUNCT fcis-3177 245 6 a. a. NOUN fcis-3177 245 7 , , PUNCT fcis-3177 245 8 bartunov bartunov PROPN fcis-3177 245 9 , , PUNCT fcis-3177 245 10 s. s. PROPN fcis-3177 245 11 , , PUNCT fcis-3177 245 12 botvinick botvinick NOUN fcis-3177 245 13 , , PUNCT fcis-3177 245 14 m. m. NOUN fcis-3177 245 15 , , PUNCT fcis-3177 245 16 wierstra wierstra PROPN fcis-3177 245 17 , , PUNCT fcis-3177 245 18 d. d. PROPN fcis-3177 245 19 , , PUNCT fcis-3177 245 20 & & CCONJ fcis-3177 245 21 lillicrap lillicrap PROPN fcis-3177 245 22 , , PUNCT fcis-3177 245 23 t. t. PROPN fcis-3177 245 24 ( ( PUNCT fcis-3177 245 25 2016 2016 NUM fcis-3177 245 26 , , PUNCT fcis-3177 245 27 june june PROPN fcis-3177 245 28 ) ) PUNCT fcis-3177 245 29 . . PUNCT fcis-3177 246 1 meta meta VERB fcis-3177 246 2 - - PUNCT fcis-3177 246 3 learning learning NOUN fcis-3177 246 4 with with ADP fcis-3177 246 5 memoryaugmented memoryaugmente VERB fcis-3177 246 6 neural neural ADJ fcis-3177 246 7 networks network NOUN fcis-3177 246 8 . . PUNCT fcis-3177 247 1 in in ADP fcis-3177 247 2 international international ADJ fcis-3177 247 3 conference conference NOUN fcis-3177 247 4 on on ADP fcis-3177 247 5 machine machine NOUN fcis-3177 247 6 learning learning NOUN fcis-3177 247 7 ( ( PUNCT fcis-3177 247 8 pp pp ADJ fcis-3177 247 9 . . PUNCT fcis-3177 247 10 1842 1842 NUM fcis-3177 247 11 - - SYM fcis-3177 247 12 1850 1850 NUM fcis-3177 247 13 ) ) PUNCT fcis-3177 247 14 . . PUNCT fcis-3177 248 1 pmlr pmlr NOUN fcis-3177 248 2 . . PUNCT fcis-3177 249 1 [ [ X fcis-3177 249 2 22 22 NUM fcis-3177 249 3 ] ] PUNCT fcis-3177 249 4 vinyals vinyal NOUN fcis-3177 249 5 , , PUNCT fcis-3177 249 6 o. o. PROPN fcis-3177 249 7 , , PUNCT fcis-3177 249 8 blundell blundell PROPN fcis-3177 249 9 , , PUNCT fcis-3177 249 10 c. c. PROPN fcis-3177 249 11 , , PUNCT fcis-3177 249 12 lillicrap lillicrap NOUN fcis-3177 249 13 , , PUNCT fcis-3177 249 14 t. t. PROPN fcis-3177 249 15 , , PUNCT fcis-3177 249 16 & & CCONJ fcis-3177 249 17 wierstra wierstra PROPN fcis-3177 249 18 , , PUNCT fcis-3177 249 19 d. d. PROPN fcis-3177 249 20 ( ( PUNCT fcis-3177 249 21 2016 2016 NUM fcis-3177 249 22 ) ) PUNCT fcis-3177 249 23 . . PUNCT fcis-3177 250 1 matching match VERB fcis-3177 250 2 networks network NOUN fcis-3177 250 3 for for ADP fcis-3177 250 4 one one NUM fcis-3177 250 5 shot shot NOUN fcis-3177 250 6 learning learning NOUN fcis-3177 250 7 . . PUNCT fcis-3177 251 1 advances advance NOUN fcis-3177 251 2 in in ADP fcis-3177 251 3 neural neural ADJ fcis-3177 251 4 information information NOUN fcis-3177 251 5 processing processing NOUN fcis-3177 251 6 systems system NOUN fcis-3177 251 7 , , PUNCT fcis-3177 251 8 29 29 NUM fcis-3177 251 9 . . PUNCT fcis-3177 252 1 [ [ X fcis-3177 252 2 23 23 NUM fcis-3177 252 3 ] ] X fcis-3177 252 4 snell snell PROPN fcis-3177 252 5 , , PUNCT fcis-3177 252 6 j. j. PROPN fcis-3177 252 7 , , PUNCT fcis-3177 252 8 swersky swersky PROPN fcis-3177 252 9 , , PUNCT fcis-3177 252 10 k. k. PROPN fcis-3177 252 11 , , PUNCT fcis-3177 252 12 & & CCONJ fcis-3177 252 13 zemel zemel PROPN fcis-3177 252 14 , , PUNCT fcis-3177 252 15 r. r. PROPN fcis-3177 252 16 ( ( PUNCT fcis-3177 252 17 2017 2017 NUM fcis-3177 252 18 ) ) PUNCT fcis-3177 252 19 . . PUNCT fcis-3177 253 1 prototypical prototypical ADJ fcis-3177 253 2 networks network NOUN fcis-3177 253 3 for for ADP fcis-3177 253 4 few few ADJ fcis-3177 253 5 - - PUNCT fcis-3177 253 6 shot shot NOUN fcis-3177 253 7 learning learning NOUN fcis-3177 253 8 . . PUNCT fcis-3177 254 1 advances advance NOUN fcis-3177 254 2 in in ADP fcis-3177 254 3 neural neural ADJ fcis-3177 254 4 information information NOUN fcis-3177 254 5 processing processing NOUN fcis-3177 254 6 systems system NOUN fcis-3177 254 7 , , PUNCT fcis-3177 254 8 30 30 NUM fcis-3177 254 9 . . PUNCT fcis-3177 255 1 [ [ X fcis-3177 255 2 24 24 NUM fcis-3177 255 3 ] ] X fcis-3177 255 4 sung sung PROPN fcis-3177 255 5 , , PUNCT fcis-3177 255 6 f. f. PROPN fcis-3177 255 7 , , PUNCT fcis-3177 255 8 yang yang PROPN fcis-3177 255 9 , , PUNCT fcis-3177 255 10 y. y. PROPN fcis-3177 255 11 , , PUNCT fcis-3177 255 12 zhang zhang PROPN fcis-3177 255 13 , , PUNCT fcis-3177 255 14 l. l. PROPN fcis-3177 255 15 , , PUNCT fcis-3177 255 16 xiang xiang PROPN fcis-3177 255 17 , , PUNCT fcis-3177 255 18 t. t. PROPN fcis-3177 255 19 , , PUNCT fcis-3177 255 20 torr torr PROPN fcis-3177 255 21 , , PUNCT fcis-3177 255 22 p. p. PROPN fcis-3177 255 23 h. h. PROPN fcis-3177 255 24 , , PUNCT fcis-3177 255 25 & & CCONJ fcis-3177 255 26 hospedales hospedale NOUN fcis-3177 255 27 , , PUNCT fcis-3177 255 28 t. t. NOUN fcis-3177 255 29 m. m. NOUN fcis-3177 255 30 ( ( PUNCT fcis-3177 255 31 2018 2018 NUM fcis-3177 255 32 ) ) PUNCT fcis-3177 255 33 . . PUNCT fcis-3177 256 1 learning learn VERB fcis-3177 256 2 to to PART fcis-3177 256 3 compare compare VERB fcis-3177 256 4 : : PUNCT fcis-3177 256 5 relation relation NOUN fcis-3177 256 6 network network NOUN fcis-3177 256 7 for for ADP fcis-3177 256 8 few few ADJ fcis-3177 256 9 - - PUNCT fcis-3177 256 10 shot shot NOUN fcis-3177 256 11 learning learning NOUN fcis-3177 256 12 . . PUNCT fcis-3177 257 1 in in ADP fcis-3177 257 2 proceedings proceeding NOUN fcis-3177 257 3 of of ADP fcis-3177 257 4 the the DET fcis-3177 257 5 ieee ieee NOUN fcis-3177 257 6 conference conference NOUN fcis-3177 257 7 on on ADP fcis-3177 257 8 computer computer NOUN fcis-3177 257 9 vision vision NOUN fcis-3177 257 10 and and CCONJ fcis-3177 257 11 pattern pattern NOUN fcis-3177 257 12 recognition recognition NOUN fcis-3177 257 13 ( ( PUNCT fcis-3177 257 14 pp pp ADJ fcis-3177 257 15 . . PUNCT fcis-3177 258 1 1199 1199 NUM fcis-3177 258 2 - - SYM fcis-3177 258 3 1208 1208 NUM fcis-3177 258 4 ) ) PUNCT fcis-3177 258 5 . . PUNCT fcis-3177 259 1 115 115 NUM fcis-3177 260 1 [ [ X fcis-3177 260 2 25 25 NUM fcis-3177 260 3 ] ] X fcis-3177 260 4 naik naik PROPN fcis-3177 260 5 , , PUNCT fcis-3177 260 6 d. d. PROPN fcis-3177 260 7 k. k. PROPN fcis-3177 260 8 , , PUNCT fcis-3177 260 9 & & CCONJ fcis-3177 260 10 mammone mammone PROPN fcis-3177 260 11 , , PUNCT fcis-3177 260 12 r. r. PROPN fcis-3177 260 13 j. j. PROPN fcis-3177 260 14 ( ( PUNCT fcis-3177 260 15 1992 1992 NUM fcis-3177 260 16 , , PUNCT fcis-3177 260 17 june june PROPN fcis-3177 260 18 ) ) PUNCT fcis-3177 260 19 . . PUNCT fcis-3177 261 1 meta meta ADJ fcis-3177 261 2 - - PUNCT fcis-3177 261 3 neural neural ADJ fcis-3177 261 4 networks network NOUN fcis-3177 261 5 that that PRON fcis-3177 261 6 learn learn VERB fcis-3177 261 7 by by ADP fcis-3177 261 8 learning learn VERB fcis-3177 261 9 . . PUNCT fcis-3177 262 1 in in ADP fcis-3177 262 2 [ [ X fcis-3177 262 3 proceedings proceeding NOUN fcis-3177 262 4 1992 1992 NUM fcis-3177 262 5 ] ] PUNCT fcis-3177 262 6 ijcnn ijcnn VERB fcis-3177 262 7 international international ADJ fcis-3177 262 8 joint joint ADJ fcis-3177 262 9 conference conference NOUN fcis-3177 262 10 on on ADP fcis-3177 262 11 neural neural ADJ fcis-3177 262 12 networks network NOUN fcis-3177 262 13 ( ( PUNCT fcis-3177 262 14 vol vol NOUN fcis-3177 262 15 . . NOUN fcis-3177 262 16 1 1 NUM fcis-3177 262 17 , , PUNCT fcis-3177 262 18 pp pp ADJ fcis-3177 262 19 . . PUNCT fcis-3177 263 1 437 437 NUM fcis-3177 263 2 - - SYM fcis-3177 263 3 442 442 NUM fcis-3177 263 4 ) ) PUNCT fcis-3177 263 5 . . PUNCT fcis-3177 264 1 ieee ieee PROPN fcis-3177 264 2 . . PUNCT fcis-3177 265 1 [ [ X fcis-3177 265 2 26 26 NUM fcis-3177 265 3 ] ] X fcis-3177 265 4 andrychowicz andrychowicz NOUN fcis-3177 265 5 , , PUNCT fcis-3177 265 6 m. m. NOUN fcis-3177 265 7 , , PUNCT fcis-3177 265 8 denil denil PROPN fcis-3177 265 9 , , PUNCT fcis-3177 265 10 m. m. NOUN fcis-3177 265 11 , , PUNCT fcis-3177 265 12 gomez gomez PROPN fcis-3177 265 13 , , PUNCT fcis-3177 265 14 s. s. PROPN fcis-3177 265 15 , , PUNCT fcis-3177 265 16 hoffman hoffman PROPN fcis-3177 265 17 , , PUNCT fcis-3177 265 18 m. m. PROPN fcis-3177 265 19 w. w. PROPN fcis-3177 265 20 , , PUNCT fcis-3177 265 21 pfau pfau PROPN fcis-3177 265 22 , , PUNCT fcis-3177 265 23 d. d. PROPN fcis-3177 265 24 , , PUNCT fcis-3177 265 25 schaul schaul PROPN fcis-3177 265 26 , , PUNCT fcis-3177 265 27 t. t. PROPN fcis-3177 265 28 , , PUNCT fcis-3177 265 29 ... ... PUNCT fcis-3177 265 30 & & CCONJ fcis-3177 265 31 de de X fcis-3177 265 32 freitas freitas PROPN fcis-3177 265 33 , , PUNCT fcis-3177 265 34 n. n. NOUN fcis-3177 265 35 ( ( PUNCT fcis-3177 265 36 2016 2016 NUM fcis-3177 265 37 ) ) PUNCT fcis-3177 265 38 . . PUNCT fcis-3177 266 1 learning learn VERB fcis-3177 266 2 to to PART fcis-3177 266 3 learn learn VERB fcis-3177 266 4 by by ADP fcis-3177 266 5 gradient gradient ADJ fcis-3177 266 6 descent descent NOUN fcis-3177 266 7 by by ADP fcis-3177 266 8 gradient gradient ADJ fcis-3177 266 9 descent descent NOUN fcis-3177 266 10 . . PUNCT fcis-3177 267 1 advances advance NOUN fcis-3177 267 2 in in ADP fcis-3177 267 3 neural neural ADJ fcis-3177 267 4 information information NOUN fcis-3177 267 5 processing processing NOUN fcis-3177 267 6 systems system NOUN fcis-3177 267 7 , , PUNCT fcis-3177 267 8 29 29 NUM fcis-3177 267 9 . . PUNCT fcis-3177 268 1 [ [ X fcis-3177 268 2 27 27 NUM fcis-3177 268 3 ] ] PUNCT fcis-3177 268 4 vinyals vinyal NOUN fcis-3177 268 5 , , PUNCT fcis-3177 268 6 o. o. PROPN fcis-3177 268 7 , , PUNCT fcis-3177 268 8 blundell blundell PROPN fcis-3177 268 9 , , PUNCT fcis-3177 268 10 c. c. PROPN fcis-3177 268 11 , , PUNCT fcis-3177 268 12 lillicrap lillicrap NOUN fcis-3177 268 13 , , PUNCT fcis-3177 268 14 t. t. PROPN fcis-3177 268 15 , , PUNCT fcis-3177 268 16 & & CCONJ fcis-3177 268 17 wierstra wierstra PROPN fcis-3177 268 18 , , PUNCT fcis-3177 268 19 d. d. PROPN fcis-3177 268 20 ( ( PUNCT fcis-3177 268 21 2016 2016 NUM fcis-3177 268 22 ) ) PUNCT fcis-3177 268 23 . . PUNCT fcis-3177 269 1 matching match VERB fcis-3177 269 2 networks network NOUN fcis-3177 269 3 for for ADP fcis-3177 269 4 one one NUM fcis-3177 269 5 shot shot NOUN fcis-3177 269 6 learning learning NOUN fcis-3177 269 7 . . PUNCT fcis-3177 270 1 advances advance NOUN fcis-3177 270 2 in in ADP fcis-3177 270 3 neural neural ADJ fcis-3177 270 4 information information NOUN fcis-3177 270 5 processing processing NOUN fcis-3177 270 6 systems system NOUN fcis-3177 270 7 , , PUNCT fcis-3177 270 8 29 29 NUM fcis-3177 270 9 . . PUNCT fcis-3177 271 1 [ [ X fcis-3177 271 2 28 28 NUM fcis-3177 271 3 ] ] X fcis-3177 271 4 lake lake NOUN fcis-3177 271 5 , , PUNCT fcis-3177 271 6 b. b. PROPN fcis-3177 271 7 m. m. PROPN fcis-3177 271 8 , , PUNCT fcis-3177 271 9 salakhutdinov salakhutdinov PROPN fcis-3177 271 10 , , PUNCT fcis-3177 271 11 r. r. PROPN fcis-3177 271 12 , , PUNCT fcis-3177 271 13 & & CCONJ fcis-3177 271 14 tenenbaum tenenbaum PROPN fcis-3177 271 15 , , PUNCT fcis-3177 271 16 j. j. PROPN fcis-3177 271 17 b. b. PROPN fcis-3177 271 18 ( ( PUNCT fcis-3177 271 19 2015 2015 NUM fcis-3177 271 20 ) ) PUNCT fcis-3177 271 21 . . PUNCT fcis-3177 272 1 human human ADJ fcis-3177 272 2 - - PUNCT fcis-3177 272 3 level level NOUN fcis-3177 272 4 concept concept NOUN fcis-3177 272 5 learning learn VERB fcis-3177 272 6 through through ADP fcis-3177 272 7 probabilistic probabilistic ADJ fcis-3177 272 8 program program NOUN fcis-3177 272 9 induction induction NOUN fcis-3177 272 10 . . PUNCT fcis-3177 273 1 science science NOUN fcis-3177 273 2 , , PUNCT fcis-3177 273 3 350(6266 350(6266 NUM fcis-3177 273 4 ) ) PUNCT fcis-3177 273 5 , , PUNCT fcis-3177 273 6 1332 1332 NUM fcis-3177 273 7 - - SYM fcis-3177 273 8 1338 1338 NUM fcis-3177 273 9 . . PUNCT fcis-3177 274 1 [ [ X fcis-3177 274 2 29 29 NUM fcis-3177 274 3 ] ] X fcis-3177 274 4 khosla khosla PROPN fcis-3177 274 5 , , PUNCT fcis-3177 274 6 a. a. NOUN fcis-3177 274 7 , , PUNCT fcis-3177 274 8 jayadevaprakash jayadevaprakash ADV fcis-3177 274 9 , , PUNCT fcis-3177 274 10 n. n. NOUN fcis-3177 274 11 , , PUNCT fcis-3177 274 12 yao yao PROPN fcis-3177 274 13 , , PUNCT fcis-3177 274 14 b. b. PROPN fcis-3177 274 15 , , PUNCT fcis-3177 274 16 & & CCONJ fcis-3177 274 17 li li PROPN fcis-3177 274 18 , , PUNCT fcis-3177 274 19 f. f. PROPN fcis-3177 274 20 f. f. PROPN fcis-3177 274 21 ( ( PUNCT fcis-3177 274 22 2011 2011 NUM fcis-3177 274 23 , , PUNCT fcis-3177 274 24 june june PROPN fcis-3177 274 25 ) ) PUNCT fcis-3177 274 26 . . PUNCT fcis-3177 275 1 novel novel NOUN fcis-3177 275 2 dataset dataset NOUN fcis-3177 275 3 for for ADP fcis-3177 275 4 fine fine ADV fcis-3177 275 5 - - PUNCT fcis-3177 275 6 grained grain VERB fcis-3177 275 7 image image NOUN fcis-3177 275 8 categorization categorization NOUN fcis-3177 275 9 : : PUNCT fcis-3177 275 10 stanford stanford PROPN fcis-3177 275 11 dogs dog NOUN fcis-3177 275 12 . . PUNCT fcis-3177 276 1 in in ADP fcis-3177 276 2 proc proc NOUN fcis-3177 276 3 . . PUNCT fcis-3177 277 1 cvpr cvpr NOUN fcis-3177 277 2 workshop workshop NOUN fcis-3177 277 3 on on ADP fcis-3177 277 4 fine fine ADV fcis-3177 277 5 - - PUNCT fcis-3177 277 6 grained grain VERB fcis-3177 277 7 visual visual ADJ fcis-3177 277 8 categorization categorization NOUN fcis-3177 277 9 ( ( PUNCT fcis-3177 277 10 fgvc fgvc NOUN fcis-3177 277 11 ) ) PUNCT fcis-3177 277 12 ( ( PUNCT fcis-3177 277 13 vol vol NOUN fcis-3177 277 14 . . NOUN fcis-3177 278 1 2 2 NUM fcis-3177 278 2 , , PUNCT fcis-3177 278 3 no no INTJ fcis-3177 278 4 . . NOUN fcis-3177 278 5 1 1 NUM fcis-3177 278 6 ) ) PUNCT fcis-3177 278 7 . . PUNCT fcis-3177 279 1 citeseer citeseer NOUN fcis-3177 279 2 . . PUNCT fcis-3177 280 1 [ [ X fcis-3177 280 2 30 30 NUM fcis-3177 280 3 ] ] X fcis-3177 280 4 han han PROPN fcis-3177 280 5 , , PUNCT fcis-3177 280 6 x. x. PROPN fcis-3177 280 7 , , PUNCT fcis-3177 280 8 zhu zhu PROPN fcis-3177 280 9 , , PUNCT fcis-3177 280 10 h. h. PROPN fcis-3177 280 11 , , PUNCT fcis-3177 280 12 yu yu PROPN fcis-3177 280 13 , , PUNCT fcis-3177 280 14 p. p. PROPN fcis-3177 280 15 , , PUNCT fcis-3177 280 16 wang wang PROPN fcis-3177 280 17 , , PUNCT fcis-3177 280 18 z. z. PROPN fcis-3177 280 19 , , PUNCT fcis-3177 280 20 yao yao PROPN fcis-3177 280 21 , , PUNCT fcis-3177 280 22 y. y. PROPN fcis-3177 280 23 , , PUNCT fcis-3177 280 24 liu liu PROPN fcis-3177 280 25 , , PUNCT fcis-3177 280 26 z. z. PROPN fcis-3177 280 27 , , PUNCT fcis-3177 280 28 & & CCONJ fcis-3177 280 29 sun sun PROPN fcis-3177 280 30 , , PUNCT fcis-3177 280 31 m. m. NOUN fcis-3177 280 32 ( ( PUNCT fcis-3177 280 33 2018 2018 NUM fcis-3177 280 34 , , PUNCT fcis-3177 280 35 january january PROPN fcis-3177 280 36 ) ) PUNCT fcis-3177 280 37 . . PUNCT fcis-3177 281 1 fewrel fewrel NOUN fcis-3177 281 2 : : PUNCT fcis-3177 281 3 a a DET fcis-3177 281 4 large large ADJ fcis-3177 281 5 - - PUNCT fcis-3177 281 6 scale scale NOUN fcis-3177 281 7 supervised supervise VERB fcis-3177 281 8 fewshot fewshot ADJ fcis-3177 281 9 relation relation NOUN fcis-3177 281 10 classification classification NOUN fcis-3177 281 11 dataset dataset VERB fcis-3177 281 12 with with ADP fcis-3177 281 13 state state NOUN fcis-3177 281 14 - - PUNCT fcis-3177 281 15 of of ADP fcis-3177 281 16 - - PUNCT fcis-3177 281 17 the the DET fcis-3177 281 18 - - PUNCT fcis-3177 281 19 art art NOUN fcis-3177 281 20 evaluation evaluation NOUN fcis-3177 281 21 . . PUNCT fcis-3177 282 1 in in ADP fcis-3177 282 2 emnlp emnlp NOUN fcis-3177 282 3 . . PUNCT fcis-3177 283 1 [ [ X fcis-3177 283 2 31 31 NUM fcis-3177 283 3 ] ] SYM fcis-3177 283 4 yu yu PROPN fcis-3177 283 5 , , PUNCT fcis-3177 283 6 m. m. NOUN fcis-3177 283 7 , , PUNCT fcis-3177 283 8 guo guo PROPN fcis-3177 283 9 , , PUNCT fcis-3177 283 10 x. x. PROPN fcis-3177 283 11 , , PUNCT fcis-3177 283 12 yi yi PROPN fcis-3177 283 13 , , PUNCT fcis-3177 283 14 j. j. PROPN fcis-3177 283 15 , , PUNCT fcis-3177 283 16 chang chang PROPN fcis-3177 283 17 , , PUNCT fcis-3177 283 18 s. s. PROPN fcis-3177 283 19 , , PUNCT fcis-3177 283 20 potdar potdar NOUN fcis-3177 283 21 , , PUNCT fcis-3177 283 22 s. s. PROPN fcis-3177 283 23 , , PUNCT fcis-3177 283 24 cheng cheng PROPN fcis-3177 283 25 , , PUNCT fcis-3177 283 26 y. y. PROPN fcis-3177 283 27 , , PUNCT fcis-3177 283 28 ... ... PUNCT fcis-3177 283 29 & & CCONJ fcis-3177 283 30 zhou zhou PROPN fcis-3177 283 31 , , PUNCT fcis-3177 283 32 b. b. PROPN fcis-3177 283 33 ( ( PUNCT fcis-3177 283 34 2018 2018 NUM fcis-3177 283 35 , , PUNCT fcis-3177 283 36 january january PROPN fcis-3177 283 37 ) ) PUNCT fcis-3177 283 38 . . PUNCT fcis-3177 284 1 diverse diverse ADJ fcis-3177 284 2 few few ADJ fcis-3177 284 3 - - PUNCT fcis-3177 284 4 shot shot NOUN fcis-3177 284 5 text text NOUN fcis-3177 284 6 classification classification NOUN fcis-3177 284 7 with with ADP fcis-3177 284 8 multiple multiple ADJ fcis-3177 284 9 metrics metric NOUN fcis-3177 284 10 . . PUNCT fcis-3177 285 1 in in ADP fcis-3177 285 2 naacl naacl PROPN fcis-3177 285 3 - - PUNCT fcis-3177 285 4 hlt hlt NOUN fcis-3177 285 5 . . PUNCT fcis-3177 286 1 [ [ X fcis-3177 286 2 32 32 NUM fcis-3177 286 3 ] ] SYM fcis-3177 286 4 geng geng PROPN fcis-3177 286 5 , , PUNCT fcis-3177 286 6 r. r. PROPN fcis-3177 286 7 , , PUNCT fcis-3177 286 8 li li PROPN fcis-3177 286 9 , , PUNCT fcis-3177 286 10 b. b. PROPN fcis-3177 286 11 , , PUNCT fcis-3177 286 12 li li PROPN fcis-3177 286 13 , , PUNCT fcis-3177 286 14 y. y. PROPN fcis-3177 286 15 , , PUNCT fcis-3177 286 16 zhu zhu PROPN fcis-3177 286 17 , , PUNCT fcis-3177 286 18 x. x. PROPN fcis-3177 286 19 , , PUNCT fcis-3177 286 20 jian jian PROPN fcis-3177 286 21 , , PUNCT fcis-3177 286 22 p. p. PROPN fcis-3177 286 23 , , PUNCT fcis-3177 286 24 & & CCONJ fcis-3177 286 25 sun sun PROPN fcis-3177 286 26 , , PUNCT fcis-3177 286 27 j. j. PROPN fcis-3177 286 28 ( ( PUNCT fcis-3177 286 29 2019 2019 NUM fcis-3177 286 30 , , PUNCT fcis-3177 286 31 november november PROPN fcis-3177 286 32 ) ) PUNCT fcis-3177 286 33 . . PUNCT fcis-3177 287 1 induction induction NOUN fcis-3177 287 2 networks network NOUN fcis-3177 287 3 for for ADP fcis-3177 287 4 few few ADJ fcis-3177 287 5 - - PUNCT fcis-3177 287 6 shot shot NOUN fcis-3177 287 7 text text NOUN fcis-3177 287 8 classification classification NOUN fcis-3177 287 9 . . PUNCT fcis-3177 288 1 in in ADP fcis-3177 288 2 proceedings proceeding NOUN fcis-3177 288 3 of of ADP fcis-3177 288 4 the the DET fcis-3177 288 5 2019 2019 NUM fcis-3177 288 6 conference conference NOUN fcis-3177 288 7 on on ADP fcis-3177 288 8 empirical empirical ADJ fcis-3177 288 9 methods method NOUN fcis-3177 288 10 in in ADP fcis-3177 288 11 natural natural ADJ fcis-3177 288 12 language language NOUN fcis-3177 288 13 processing processing NOUN fcis-3177 288 14 and and CCONJ fcis-3177 288 15 the the DET fcis-3177 288 16 9th 9th ADJ fcis-3177 288 17 international international ADJ fcis-3177 288 18 joint joint ADJ fcis-3177 288 19 conference conference NOUN fcis-3177 288 20 on on ADP fcis-3177 288 21 natural natural ADJ fcis-3177 288 22 language language NOUN fcis-3177 288 23 processing processing NOUN fcis-3177 288 24 ( ( PUNCT fcis-3177 288 25 emnlp emnlp NOUN fcis-3177 288 26 - - PUNCT fcis-3177 288 27 ijcnlp ijcnlp NOUN fcis-3177 288 28 ) ) PUNCT fcis-3177 288 29 ( ( PUNCT fcis-3177 288 30 pp pp ADJ fcis-3177 288 31 . . PUNCT fcis-3177 289 1 3904 3904 NUM fcis-3177 289 2 - - SYM fcis-3177 289 3 3913 3913 NUM fcis-3177 289 4 ) ) PUNCT fcis-3177 289 5 . . PUNCT