Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 16 https://internationalpubls.com Nonlinear Dynamics of Neural Networks: Applications in Pattern Recognition Dr. F Rahman Assistant Professor, Faculty of CS & IT, Kalinga University, Naya Raipur, Chhattisgarh, India. Mail ID:ku.frahman@kalingauniversity.ac.in Article History: Received: 22-05-2023 Revised: 28-06-2023 Accepted: 02-08-2023 Abstract: Improvements in pattern recognition systems are best achieved by fully comprehending and utilizing the nonlinear dynamics of neural networks. When dealing with complex, nonlinear patterns in real-world data, a nuanced approach is necessary to fully utilize neural networks' potential. By improving performance, interpretability, and flexibility in complicated pattern recognition tasks, Nonlinear Dynamics method meet this demand. Problems with interpretability, complicated training, and generalization in the presence of noisy data are all aspects of nonlinear dynamics that pose difficulties. These complex patterns are difficult for traditional neural networks to capture and comprehend. In response to these difficulties, Nonlinear Dynamics has been created to offer a comprehensive approach to making good use of nonlinear dynamics in pattern recognition. This research presents a novel method for dealing with problems caused by nonlinear dynamics, called Nonlinear Dynamics-Driven Adaptive Neural Network (ND-ANN). To achieve the sweet spot between model complexity and interpretability, NDANN employs a hybrid learning algorithm (HLA), an explainability module, ensemble integration, regularization for stability, and an adaptive architecture. This novel approach lays the groundwork for future developments in pattern recognition technology and guarantees better performance when capturing complicated patterns. Improved pattern recognition systems made possible by NDANN's accurate modeling of nonlinear dynamics pave the way for new developments in healthcare, information processing, and technology. Validation of NDANN's effectiveness in dealing with nonlinear dynamics issues was achieved through extensive simulation analyses. Precision, accuracy, recall, and F1 score are some of the performance indicators that undergo thorough evaluation across various datasets. The inclusion of nonlinear dynamics in neural networks has the ability to transform pattern recognition, and the simulation results show that NDANN is better than traditional models. Keywords: Nonlinear, Dynamics, Neural Networks, Pattern Recognition, Adaptive. 1. Introduction Emerging issues with "Nonlinear Dynamics of Neural Networks: Applications in Pattern Recognition" are complex and affect how well neural networks operate in practice [1]. Although there are challenges to implementing neural networks in the real world, their nonlinear dynamics show great potential for capturing intricate patterns and correlations [2]. A major issue is how well the network's learnt representations can be explained and understood [3]. Important for applications like pattern recognition in delicate domains, nonlinear dynamics can cause complex and abstract modifications of incoming data, making it hard to understand the logic behind network decisions [4]. Overfitting and vanishing gradients are problems that can arise during the optimization of very nonlinear architectures during training, which can reduce the model's ability to generalize to new Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 17 https://internationalpubls.com data [5]. Another obstacle is the computational and scalability requirements of complicated nonlinear dynamics, which are particularly problematic for large-scale applications that necessitate processing in real-time [6]. Due to the fact that even small changes in input can cause misclassifications [7], nonlinear neural networks are becoming increasingly susceptible to adversarial attacks [8]. Conducting comprehensive research into interpretability, optimization strategies, scalability, and security is essential for bridging the gap between theoretical advancements in nonlinear dynamics and their practical implementation for robust pattern recognition [9]. This will ensure that the potential benefits of nonlinear neural networks in pattern recognition are responsibly and effectively realized [10]. In the field of "Nonlinear Dynamics of Neural Networks: Applications in Pattern Recognition," current methods frequently employ complicated, nonlinear structures to extract detailed correlations and patterns from data [11]. To simulate the complexity of real-world data, methods like convolutional neural networks (CNNs), recurrent neural networks (RNNs), and deep neural networks (DNNs) use complicated connection patterns and nonlinear activation functions [12]. Techniques like reservoir computing use recurrent neural networks, which are dynamic, to process data over time [13]. Applying nonlinear dynamics to pattern recognition is still not without its hurdles, even with current developments [14]. They are not yet widely used in mission-critical applications due to concerns about interpretability, which arises from the fact that these networks undergo complex transformations that make it hard to comprehend and explain the decision-making process [15]. Overfitting, dealing with vanishing gradients, and guaranteeing robust generalization are only a few of the difficulties that arise during the training of these nonlinear models. The computational needs of complex systems pose a threat to scalability, which in turn affects real-time processing capabilities, especially in contexts with limited resources [16]. There are security concerns with using nonlinear neural networks for pattern recognition because of their susceptibility to adversarial assaults. To overcome these obstacles, researchers must keep working to make neural networks more interpretable, optimize training procedures, make them more scalable, and strengthen the security of nonlinear dynamics that they can recognize patterns effectively and reliably. β€’ By capitalizing on a thorough comprehension of the nonlinear dynamics present in neural networks, the major goal of the research is to improve pattern recognition systems. The goal is to discover how to fully utilize neural networks to capture and understand complicated, nonlinear patterns in real-world data, which can be quite challenging. The primary goals of the research are to develop more efficient and adaptable systems that can handle complex pattern recognition jobs with more ease. β€’ The present research presents a new approach to pattern recognition problem solving by introducing the Nonlinear Dynamics-Driven Adaptive Neural Network (NDANN). NDANN uses an adaptive architecture, regularization for stability, an ensemble integration, an explainability module, and a hybrid learning method. Finding a happy medium between model complexity and interpretability is the goal, with the end goal being an adaptive solution that solves problems with generalization, training complexity, and interpretability while dealing with noisy data. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 18 https://internationalpubls.com β€’ This research endeavors to undertake thorough simulation evaluations in order to establish NDANN as a standard for handling nonlinear dynamics concerns. There is extensive testing across several datasets for performance metrics like F1 score, recall, accuracy, and precision. The goal is to show that NDANN and other neural networks with nonlinear dynamics can improve pattern recognition systems over traditional models in terms of accuracy and efficiency. By paving the way for better and more precise pattern recognition systems, this study paves the way for improvements in healthcare, data processing, and technology. The present research follows a format that is consistent with the literature review given in Section 2: Nonlinear Dynamics of Neural Networks. The mathematical foundations of ND-ANN, or dynamics- driven adaptive neural networks, are laid out in Section 3. The research findings and discussion are presented in Section 4, while the summary and concluding remarks are presented in Section 5. 2. Literature Survey This review explores recent developments in various AI applications using artificial neural networks and computational methods. The authors Abiodun et al. discuss difficulties with artificial neural networks (ANNs) used for pattern recognition and introduce AI methods for finding nonlinear dynamical systems. Issues with applying ANNs to pattern recognition (PR) [17] are discussed by Abiodun, O. I. et al. The report highlights problems that are preventing research from progressing, such as a lack of literature and a whimsical perspective. The paper reviews existing ANN models and demonstrates how well they perform PR tasks. Their outstanding performance is supported by statistical indicators, which suggest that both existing and new models should be further prioritized for future success. Identifying nonlinear dynamical systems is made easier with the thorough overview of computational techniques in artificial intelligence (CT-AI) [18] provided by Quaranta, G et al. Parametric methods, such as particle swarm optimization and genetic algorithms, tackle models such as Bouc-Wen and chaotic systems. One can go over some nonparametric approaches that make use of ANNs and genetic programming. Demonstrating its usefulness in engineering and structural mechanics, the work lays the groundwork for related future studies. For unknown continuous-time affine nonlinear systems with actuator defects, Lin et al. offer a data- based fault-tolerant control (FTC) [19] technique. The system dynamics are modeled using a neural network (NN) identifier based on particle swarm optimization (PSO), and the Hamilton-Jacobi- Bellman equation is efficiently solved using a PSO-optimized critic neural network (PSOCNN). Validated by simulations showing its efficacy, the suggested FTC method guarantees stability under actuator defects. To forecast complicated nonlinear pulse propagation in optical fibers using only input pulse intensity profiles, Salmela, L. et al. present a recurrent neural network (RNN) [20]. The method's predictions meet experimental data in a good way, as seen in pulse compression and supercontinuum creation. With its adaptability to different input conditions and fiber systems, this method presents a great opportunity for optimizing and designing experiments in photonic technologies in real-time. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 19 https://internationalpubls.com A supervised Long Short-Term Memory (SLSTM) [21] network is presented by Yuan, X et al., for use with soft sensors in manufacturing processes. To learn dynamic hidden states, SLSTM takes into account both input and quality variables, which improves the accuracy of quality predictions compared to basic LSTM units. People show that SLSTM can effectively capture dynamic and nonlinear process behaviours on penicillin fermentation and an industrial debutanizer column. Finally, these experiments show that different AI techniques can handle different kinds of problems in different ways. With its exceptional performance in several applications, the Nonlinear Dynamics- Driven Adaptive Neural Network (ND-ANN) stands out as a formidable rival among the many available technologies, each method has its own benefits. 3. Proposed method Pattern recognition is a rapidly growing area that continually searches for new ways to deal with complex real-world data. The Nonlinear Dynamics-Driven Adaptive Neural Network (ND-ANN) is a revolutionary new approach to neural network training that overcomes the shortcomings of conventional neural networks in the capture of complicated, nonlinear patterns. By using a hybrid learning system, an explainability module, ensembles integration, and adaptive design, NDANN establishes a compromise among model complexity & interpretability. When it comes to pattern recognition, NDANN claims to be superior since it tackles issues of interpretability, training difficulty, and noise generalization. The effectiveness of NDANN has been validated by extensive simulation evaluations, demonstrating its potential to transform healthcare, processing of information, and technology by accurately simulating nonlinear dynamics. Figure 1: Layers in Nonlinear Dynamics of Neural Networks in Pattern Recognition An innovative approach in Artificial Intelligence and Machine Learning, the combination of nonlinear behaviour with neural networks as the basis has shown great promise, especially for pattern recognition. In order to improve neural networks' capacity to understand intricate data patterns and correlations, this multidisciplinary strategy integrates concepts from chaos theory, dynamical systems theory, and bifurcation analysis. The main parts and steps of this procedure are shown in Figure 1. Data in its rawest form, which may be pictures, signals, or anything else, is where the adventure starts. Everything else that happens later on in the system is built on top of this data. When it comes to improving and refining the raw input data, the Pre-processing Layer is crucial. For Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 20 https://internationalpubls.com the neural network to get clean and uniform input, activities including data augmentation, normalization, and noise reduction may be necessary. By utilizing the Feature Extraction Layer, the pre-processed data's most important and distinguishing features may be extracted. Finding and extracting key elements that augment comprehension of the input sequences is what this stage is all about. An essential part of this is using nonlinear dynamics to find complex patterns that linear approaches could miss. The layer dealing with Nonlinear Behaviour of Neural Networks is crucial to the system. This includes incorporating modern facilities ideas into the neural network design from fields including bifurcation analysis, dynamical systems theory, and chaos theory. It is now considered the network as an evolving system that may display complicated behaviours as time goes on, rather than a fixed object. A neural network may learn and adapt with the help of nonlinear dynamics, which captures the nonlinearity and temporal relationships found in many datasets from the real world. Because chaos theory is sensitive to starting circumstances, the network may learn from a wide variety of patterns and configurations. Contrarily, bifurcation analysis sheds light on state transitions by pinpointing important moments in the network's behaviour. Prior to the Classification Layer, important characteristics are extracted and nonlinear dynamics are injected. In this stage, the input data is recognized and classified using the learnt patterns. It is then assigned to predetermined classes. Applications in dynamic circumstances benefit greatly from the network's enhanced capacity to handle complex and developing patterns, which is achieved by the nonlinear dynamics layer's dynamic flexibility. The pattern recognition process's end result is the system's ultimate output. It might take the form of output that is particular to the application, such as class labels or probabilities. Data undergoes a complex journey through a neural network that is enhanced by nonlinear dynamics principles, as demonstrated in Figure 1. Improved pattern recognition and the neural network's flexibility for practical use are both achieved by using a holistic approach. This approach is a powerful technique for improving neural networks' performance in pattern recognition applications because it combines preliminary processing, extraction of features, and nonlinear dynamics. The complex interaction between activations & feedback inhibition leads to improved performance and flexibility, as shown in Figure 1, which demonstrates the nonlinear behaviour of neural networks used in recognizing patterns. 𝑄 = ∫ π‘“βˆ’(𝛽.𝑦2).cos(𝛾.𝑦).𝑄(π‘‡π‘Ÿπ‘’π‘’ π‘ƒπ‘œπ‘ π‘–π‘‘π‘–π‘£π‘’π‘ |𝑦)𝑑𝑦 ∞ βˆ’βˆž √∫ π‘“βˆ’(𝛿.𝑦2).𝑄(πΉπ‘Žπ‘™π‘ π‘’ π‘ƒπ‘œπ‘ π‘–π‘‘π‘–π‘£π‘’π‘ |𝑦)𝑑𝑦+𝛼 ∞ βˆ’βˆž (1) The total accuracy in pattern recognition is represented by 𝑄 in the equation (1), which is an expression of integrals across the whole real number line. A Gaussian distribution with parameter 𝛿 is involved in the equation (1) π‘“βˆ’(𝛿.𝑦2). 𝑄(πΉπ‘Žπ‘™π‘ π‘’ π‘ƒπ‘œπ‘ π‘–π‘‘π‘–π‘£π‘’π‘ |𝑦), which represents the probability density of false positives conditional on the continuous variable 𝑦. The impact of false positives on the total accuracy is taken into consideration in the first integral. An additional, more intricate structure is introduced by the second term, which are the product of a Gaussian distribution with parameters 𝛽 and a cosine function with parameters 𝛾 and the integral of π‘“βˆ’(𝛽.𝑦2). cos(𝛾. 𝑦) . 𝑄(π‘‡π‘Ÿπ‘’π‘’ π‘ƒπ‘œπ‘ π‘–π‘‘π‘–π‘£π‘’π‘ |𝑦). When analyzing the probability distribution of true positives across the continuous variable 𝑦, this term quantifies their effect 𝑄(π‘‡π‘Ÿπ‘’π‘’ π‘ƒπ‘œπ‘ π‘–π‘‘π‘–π‘£π‘’π‘ |𝑦). Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 21 https://internationalpubls.com The complex link between both true and false positives throughout the continuous range of the variable 𝑦 is considered in the evaluation of accuracy through the interaction of these integrals with non-linear functions, which represent a probabilistic method. 𝐡 = ∫ π‘“βˆ’(𝛽.𝑦2) βˆšπœ‹ ∞ βˆ’βˆž .π‘™π‘œπ‘”π›Ύ(𝛿.cosh(𝛼.𝑦)).𝑄(πΆπ‘œπ‘Ÿπ‘Ÿπ‘’π‘π‘‘ πΆπ‘™π‘Žπ‘ π‘ π‘–π‘“π‘–π‘π‘Žπ‘‘π‘–π‘œπ‘›|𝑦)𝑑𝑦 ∫ π‘“βˆ’(πœ–.𝑦2) βˆšπœ‹ ∞ βˆ’βˆž .π‘™π‘œπ‘”πœ—(πœ”.sinh(πœƒ.𝑦)).𝑄(πΌπ‘›π‘π‘œπ‘Ÿπ‘Ÿπ‘’π‘π‘‘ πΆπ‘™π‘Žπ‘ π‘ π‘–π‘“π‘–π‘π‘Žπ‘‘π‘–π‘œπ‘›|𝑦)𝑑𝑦+𝐿 (2) A complex measure of the overall accuracy of the model in the classification context, 𝐡 in the equation (2) is defined as the sum of all integrals across the whole real number line. The probability density of incorrect classifications conditional on the continuous variable 𝑦, as captured by the initial term, π‘“βˆ’(πœ–.𝑦2) βˆšπœ‹ . π‘™π‘œπ‘”πœ—(πœ”. sinh(πœƒ. 𝑦)). 𝑄(πΌπ‘›π‘π‘œπ‘Ÿπ‘Ÿπ‘’π‘π‘‘ πΆπ‘™π‘Žπ‘ π‘ π‘–π‘“π‘–π‘π‘Žπ‘‘π‘–π‘œπ‘›|𝑦), includes a Gaussian distribution with parameter πœ– and a logarithmic term involving a hyperbolic sine function. This metric measures how much inaccurate categorizations impacted the total precision. An additional level of complexity is introduced by the second term, which is π‘“βˆ’(𝛽.𝑦2) βˆšπœ‹ . π‘™π‘œπ‘”π›Ύ(𝛿. cosh(𝛼. 𝑦)). 𝑄(πΆπ‘œπ‘Ÿπ‘Ÿπ‘’π‘π‘‘ πΆπ‘™π‘Žπ‘ π‘ π‘–π‘“π‘–π‘π‘Žπ‘‘π‘–π‘œπ‘›|𝑦), which uses a hyperbolic cosine function and a logarithmic term. The impact of accurate classifications is taken into consideration by this term, which takes into account the probability distribution of 𝑄(πΆπ‘œπ‘Ÿπ‘Ÿπ‘’π‘π‘‘ πΆπ‘™π‘Žπ‘ π‘ π‘–π‘“π‘–π‘π‘Žπ‘‘π‘–π‘œπ‘›|𝑦) over the continuous variable 𝑦. The equation (2) represents a probabilistic and complex assessment of the model's correctness; it captures the interaction between right and wrong classifications throughout a continuous spectrum and allows for parameters to be adjusted accounting for different parts of the model's performance. 𝑆 = ∫ π‘“βˆ’(𝛽.𝑦2) βˆšπœ‹ .π‘™π‘œπ‘”π›Ύ(𝛿.arccsc(𝛼.𝑦)).𝑄(π‘‡π‘Ÿπ‘’π‘’ π‘ƒπ‘œπ‘ π‘–π‘‘π‘–π‘£π‘’π‘ |𝑦)𝑑𝑦 ∞ βˆ’βˆž √∫ π‘“βˆ’(πœ–.𝑦2) βˆšπœ‹ ∞ βˆ’βˆž .π‘™π‘œπ‘”πœ—(πœ”.arccoth(πœƒ.𝑦)).𝑄(πΉπ‘Žπ‘™π‘ π‘’ π‘π‘’π‘”π‘Žπ‘‘π‘–π‘£π‘’π‘ |𝑦)𝑑𝑦 3 +𝐿 (3) In the recall equation (3), denoted as 𝑆, the parameters (𝛽, 𝛾, 𝛼, 𝛿, πœ—, πœ”, πœƒ and 𝐿) are pivotal in customizing the analysis to certain data properties. The effect of the input variable 𝑦 on false negatives is represented by the parameter πœ– in the exponential term, which is a Gaussian distribution in the first integral. The hyperbolic arccoth functional (πœ”. arccoth(πœƒ. 𝑦)) is given by the logarithmic function with base πœ—, which captures complicated relationships in the recall evaluation and adds another degree of complexity. The analysis is further refined in its probabilistic character by 𝑄(πΉπ‘Žπ‘™π‘ π‘’ π‘π‘’π‘”π‘Žπ‘‘π‘–π‘£π‘’π‘ |𝑦). In the same way, the second integral represents the contribution of true positives and uses a second Gaussian distribution with an adjustable parameter 𝛽 and a logarithmic function with a base of 𝛾 performed to the inverse trigonometry function (𝛿. arccsc(𝛼. 𝑦)). The probability estimation is refined by 𝑄(π‘‡π‘Ÿπ‘’π‘’ π‘ƒπ‘œπ‘ π‘–π‘‘π‘–π‘£π‘’π‘ |𝑦). With the parameter 𝐿 scaling the total contribution, the recall equation (3) may be adjusted to some extent, transforming it into a flexible and advanced tool for studying the relationship among true positives & false negatives in various data situations. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 22 https://internationalpubls.com Figure 2: Architecture of Adaptive Neural Networks for Pattern Recognition Based on Nonlinear Dynamics (NDANN) The NDANN architecture, shown in Figure 2, revolutionizes pattern recognition by utilizing the combined strengths of its parts. By utilizing the Hybrid Training algorithms, Explainability Module, Ensemble integrating, Regularization for Stability, & Adaptive Architecture, NDANN ensures accurate and dependable pattern recognition while maintaining a delicate balance between complexity and interpretability. Medical treatment, data processing, and technological innovation stand to benefit greatly from this adaptable neural network. Pattern recognition using a Nonlinear Dynamics-Driven Adaptive Neural Network (NDANN) is depicted in the figure 2, which serves to explain its overall design. When conventional neural networks fail to adequately capture complicated, nonlinear patterns in actual data, NDANN steps in as a fresh, innovative answer. The Input Data is the starting point; it stands for the unprocessed data from which the algorithm will attempt to extract patterns. This data might be diverse and come from a variety of sources, including photographs, text, or signals, among others. System implementation of feature extraction and pre- processing procedures improves input data quality and relevance. Data cleansing and transformation, feature extraction, and preparation for future analysis are all part of this phase. The NDANN, a particular neural network built to properly embrace and use nonlinear dynamics, is the foundation of the system. To improve efficiency, interpretability, and adaptability in identifying complex patterns, NDANN uses an advanced approach that combines several critical components, setting it apart from conventional neural networks. By using a Hybrid Learning Algorithm, NDANN is able to achieve an appropriate balance between the model's complexity and its interpretability. As a result of this algorithm's integration of many learning strategies, the neural network can adapt to nonlinear pattern complexity while keeping its decision-making process transparent. With the Explainability Module, NDANN improves interpretability. To make the neural network's process of decision-making more clear and comprehensible, this module offers information. There are some uses where the model's interpretability is more essential than its accuracy, and this is a must. Incorporating an Ensemble Learning strategy, NDANN enhances generalizability and endurance. Ensembles are made up of Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 23 https://internationalpubls.com many models that work together to aid in recognition. Particularly when dealing with chaotic information, this guarantees more trustworthy outcomes. As a solution to the problems that arise from complex training, NDANN uses Regularization for Stability. To improve the network's generalizability to novel, unknown data, this method aids in avoiding over fitting. The structure of NDANN is dynamically adjusted by its Adaptive Architecture according to the input data's properties. The versatility of the network makes it ideal for many uses, as it improves its pattern-capturing and -representation capabilities. As a last step, the system generates the Recognition Results, which show the outcome of the pattern recognition process. By utilizing its advanced architecture, NDANN enhances and optimizes these results, demonstrating its ability to capture and comprehend subtle patterns in real-world information. Figure 2 shows how the NDANN design combines different parts to make a formidable adaptive pattern recognition system. The adaptive architecture, explainability module, ensemble integrating, regularization for stability, and hybrid learning algorithm all work together to improve pattern recognition performance in various domains by overcoming the difficulties of nonlinear dynamics. 𝐹1 = ∫ π‘“βˆ’(𝛽.𝑦2).tan(𝛾.𝑦).𝑄(π‘‡π‘Ÿπ‘’π‘’ π‘ƒπ‘œπ‘ π‘–π‘‘π‘–π‘£π‘’π‘ |𝑦)𝑑𝑦 ∞ βˆ’βˆž √∫ π‘“βˆ’(𝛿.𝑦2).cot(𝛼.𝑦).𝑄(πΉπ‘Žπ‘™π‘ π‘’ π‘ƒπ‘œπ‘ π‘–π‘‘π‘–π‘£π‘’π‘ |𝑦)𝑑𝑦 ∞ βˆ’βˆž + √∫ π‘“βˆ’(πœ–.𝑦2).𝑄(πΉπ‘Žπ‘™π‘ π‘’ π‘π‘’π‘”π‘Žπ‘‘π‘–π‘£π‘’π‘ |𝑦)𝑑𝑦 ∞ βˆ’βˆž 3 +πœ— (4) The equation (4) is an ND-ANN (Non-Deterministic Artificial Neural Network) framework-specific complicated integral equation that represents the F1-score. The integrals cover the whole real line, and the variable 𝑦 stands for the input space. The terms related to Gaussian functions, which impact the distribution of false positives & false negatives, include 𝛿, 𝛼, and πœ–. The hyperbolic & tangent functions are connected to the parameters 𝛽 and 𝛾, which help in determining the genuine positives. Parameters such as 𝛽, 𝛾, 𝛿, 𝛼, πœ–, and πœ— can be adjusted to make the equation work better in certain situations. A further level of complexity is introduced to the evaluation of true positives & false positives by the cotangent & tangent functions, cot(𝛼. 𝑦) and tan(𝛾. 𝑦), which bring trigonometric components. The input 𝑦 determines the conditional probabilities of false positives 𝑄(πΉπ‘Žπ‘™π‘ π‘’ π‘ƒπ‘œπ‘ π‘–π‘‘π‘–π‘£π‘’π‘ |𝑦) and false negatives 𝑄(πΉπ‘Žπ‘™π‘ π‘’ π‘π‘’π‘”π‘Žπ‘‘π‘–π‘£π‘’π‘ |𝑦). When applied to the ND-ANN framework, the equation (4) gives an intricate and complex evaluation of the F1-score by taking into account all the variables that go into a machine learning model's recall-to-precision ratio. π‘Šπ‘—(𝑒 + βˆ†π‘’) = π‘Šπ‘—(𝑒) + 𝛽. tanh (𝛾. π‘Šπ‘—(𝑒)) + 𝛿. sin(π‘₯𝑒 + πœ‘) + πœ”. 𝐽𝑗(𝑒) (5) In an ND-ANN, the equation (5) signifies the fluctuation of a spiking neuron's membrane potential. Neuron 𝑗 membrane potential at time 𝑒 is represented by π‘Šπ‘—(𝑒). The nonlinearity's inclination is determined by 𝛾, which is a scaling factor that affects the effect of the hyperbolic tangential nonlinearity, which is provided by tanh (𝛾. π‘Šπ‘—(𝑒)). A sinusoidal component is added to the potential of the membrane by the second term, 𝛿. sin(π‘₯𝑒 + πœ‘), where 𝛿 is the frequency, π‘₯ is the amplitude, and πœ‘ is the phase shift. The last element, πœ”. 𝐽𝑗(𝑒), takes into consideration the impact of the input current 𝐽𝑗(𝑒) on the potential of the membrane, with πœ” standing for the intensity of this impact. All of these parts work together to provide a detailed model of ND-ANN neuronal function, one that includes nonlinearities & adaptive integration driven by externally generated currents. An artificial Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 24 https://internationalpubls.com neural network can accurately portray a neuron's behaviour with finer details when the 𝛽, 𝛾, 𝛿, πœ‘, πœ” and 𝛼 are used. πœ”π‘’ = πœ”0 (1+𝛽.𝑒) + 𝛾. sin(𝛿. 𝑒 + πœ‘) + 𝛼 √1+πœ–.𝑒 (6) In ND-ANN using a chaotic attractor, the equation (6) specifies a rate of adaptive learning πœ”π‘’. The πœ”0 (1+𝛽.𝑒) decays with time, 𝛾. sin(𝛿. 𝑒 + πœ‘) is chaotic, and 𝛼 √1+πœ–.𝑒 is a chaotic attractor that decays slowly. A dynamic and intricate learning rate landscape is produced by the interplay of these components. Figure 3: Dynamic Nonlinearity in Pattern Recognition Neural Networks Figure 3 is an all-inclusive representation of the main parts and steps of using neural networks' nonlinear dynamics for comprehensive pattern recognition. In particular, when dealing with complicated, nonlinear patterns found in real-world data, the figure 3 presents the essence of suggested method, Nonlinear Dynamics of Neural Networks (NDANN), along with its influence on enhancing the performance, interpretability, and adaptability of pattern recognition systems. Nonlinear Dynamics of Neural Networks is the central idea depicted as a block in Figure 3. For complex patterns that are difficult for conventional neural networks to understand, NDANN provides a solid basis. A new world of potential in pattern recognition technologies is unlocked by NDANN as it explores the complex interaction of nonlinear dynamics in neural networks. Important performance indicators that surround NDANN and are critical for determining the efficacy of the proposed method surround it. Important metrics that are evaluated extensively across different datasets are precision, accuracy, recall, & F1 score. These indicators may be used to evaluate how well NDANN handles complicated patterns and how reliable it is. The Simulation Analyses section, which displays the thorough evaluation procedure used to confirm NDANN's effectiveness, is adjacent to the effectiveness indicators. Researchers validate that NDANN beats conventional models on all metrics (precision, accuracy, recall, & F1 score) by conducting thorough simulation studies on a variety of datasets. This part is essential for showing that NDANN is better at dealing with nonlinear behaviour in pattern recognition and that it can be used in the real world. Figure 3 covers the more general effects of NDANN's work on pattern recognition. "Future Development in Pattern Recognition" indicates, this approach has the potential to dramatically alter the field. The use of NDANN opens up new possibilities in medicine, data processing, and Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 25 https://internationalpubls.com technology by precisely simulating neural networks' nonlinear dynamics. Incorporating nonlinear dynamics into pattern recognition has exciting prospects, as shown in figure. The indicated NDANN technique is outlined in Figure, which shows the complex interplay between increased pattern recognition, neural networks, and nonlinear dynamics. The method's robustness is emphasized by the performance metrics and simulation studies, and the prospective applications and larger implications of NDANN in varied disciplines are emphasized by the perspective for future advances. It's called "Nonlinear Dynamics in Neural Networks for Pattern Recognition," and it enables to understand the way this novel approach affects the future of pattern recognition technology. The complex interplay of accuracy, precision, and the detailed assessment of NDANN is captured in Figure 3 within this framework. It graphically demonstrates the method's dependability across varied datasets by putting focus on simulation analyses. The complex nonlinear dynamics of pattern recognition neural networks are shown in Figure 3. To further understand how these dynamics lead to adaptation and improved performance in capturing complicated patterns, the neuronal ensemble's feedback inhibition mechanisms and complex activation processes are visualized in the graph. A better grasp of the network's pattern recognition and adaptation capabilities may be gained from the visual depiction, which improves comprehension of the interaction among individual forecasts, sigmoidal terms, sinusoidal elements, as well as feedback inhibition. βˆ†π‘₯π‘—π‘˜(𝑒) = 𝛽. tanh (𝛾. π‘₯π‘—π‘˜(𝑒)) + 𝛿. sin(π‘₯𝑒 + πœ‘) . π‘₯π‘—π‘˜(𝑒) + 𝛼. ( 1 1+𝑓 βˆ’πœ–.π‘₯π‘—π‘˜(𝑒) βˆ’ 1 2 ) (7) The equation (7) represents the variation in synaptic weights βˆ†π‘₯π‘—π‘˜(𝑒) in a Neural Dynamic-Adaptive Neural Network (ND-ANN). It includes a sinusoidal term 𝛿. sin(π‘₯𝑒 + πœ‘) . π‘₯π‘—π‘˜(𝑒) that introduces oscillatory behaviour, a hyperbolic tangent nonlinearity 𝛽. tanh (𝛾. π‘₯π‘—π‘˜(𝑒)), which represents adaptive influences, and a homeostatic component that modulates weights according to a sigmoidal function 𝛼. ( 1 1+𝑓 βˆ’πœ–.π‘₯π‘—π‘˜(𝑒) βˆ’ 1 2 ), avoiding growth that is unregulated The factors 𝛽, 𝛿, 𝛼, πœ–, 𝛾 and πœ‘ the size, form, and frequency of these components, which together represent the complex dynamics of synaptic plasticity. This is essential for the flexible and consistent operation of neural networks. π‘§π‘’π‘›π‘ π‘’π‘šπ‘π‘™π‘’(𝑦) = 1 𝑁 βˆ‘ ( 1 1+π‘“βˆ’π›½.𝑧𝑛(𝑦) + 𝛾. sin (𝛿. 𝑧𝑛(𝑦)))𝑁 𝑛=1 βˆ’ 𝛼. ( 1 1+π‘“βˆ’πœ–.π‘¦π‘’π‘›π‘ π‘’π‘šπ‘π‘™π‘’(𝑦)) (8) The result of the ND-ANN ensemble, characterized by a merger of individual predictions of 𝑁 neural networks, is represented by π‘§π‘’π‘›π‘ π‘’π‘šπ‘π‘™π‘’(𝑦) in the equation (8). Every single forecast 𝑧𝑛(𝑦) goes through an intricate activation procedure that incorporates a sinusoidal element with parameter 𝛼, a sigmoidal term with values 𝛾 and 𝛽, and so on. The total of these activated predictions is used to generate the ensemble average, with the inhibitory term being guided by a sigmoidal function with parameters πœ– and 𝛼. The dynamic inhibitory mechanism introduced by this feedback inhibitory term enhances the ensemble's flexibility and performance as a whole by dampening particular responses. The equation (8) effectively captures the complicated data patterns by integrating individual neural network predictions, an advanced activation function, and a feedback inhibition mechanism. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 26 https://internationalpubls.com The ND-ANN, or Nonlinear Dynamics-Driven Adaptive Neural Network, is a method for better pattern identification that the paper introduces. In order to overcome the shortcomings of conventional neural networks when it comes to understanding intricate, nonlinear patterns, ND-ANN utilizes an innovative combination of a hybrid learning approach, explainability part, ensemble integration, & adaptive design. Overcoming difficulties with interpretability, training complexity, & noise-induced generalization, ND-ANN achieves a critical equilibrium among model complexity and interpretability. Comprehensive simulation studies confirm its effectiveness, showing how ND-ANN can capture and understand complex nonlinear patterns in real-world data, which might lead to a revolution in healthcare accuracy, processing of information, and technology. 4. Results and Discussion This research examines ND-ANN, or Nonlinear Dynamics-Driven Adaptive Neural Network, and its capabilities in pattern recognition by testing it in a number of different setups and conditions. For a thorough grasp of ND-ANN's pattern detection efficacy, people investigate precision, accuracy, recall, and F1 score measures. Figure 4(a): Precision Analysis is compared with ND-ANN Figure 4(b): Precision Analysis is compared with HLA The dependability and accuracy of the Nonlinear Dynamics-Driven Adaptive Neural Network (NDANN) in capturing complicated patterns across varied datasets are revealed by its precision analysis inside the pattern recognition context. The ratio of successfully identified positive occurrences to the total anticipated positive instances is measured by precision, a crucial parameter in pattern recognition. By reducing the number of false positives, NDANN's precision analysis highlights its keen ability to detect relevant patterns. By improving its capacity to differentiate meaningful patterns from noise, NDANN's hybrid learning algorithm, explainability module, ensemble integration, regularization for stability, and adaptive architecture all work together to increase its accuracy. The efficiency of NDANN in accurately finding and classifying positive cases inside complex and dynamic patterns is confirmed by its continuously high precision levels, as shown in comprehensive simulation analyses across different datasets. Healthcare diagnostics, financial fraud detection, and security surveillance are only a few examples of real-world applications where NDANN's practical value is shown by the precise analysis. To its exceptional accuracy, NDANN is a strong and trustworthy instrument that could significantly improve pattern recognition technology in practical settings. The Nonlinear Dynamics-Driven Adaptive Neural 0 50 100 150 200 250 300 350 400 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0 P R E C IS IO N A N A L Y S IS R A T IO ( % ) NUMBER OF SAMPLES PR CT-AI FTC RNN ND-ANN 0 50 100 150 200 250 300 350 400 450 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0 P R E C IS IO N A N A L Y S IS R A T IO ( % ) NUMBER OF SAMPLES PR CT-AI FTC RNN HLA Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 27 https://internationalpubls.com Network (ND-ANN) achieved an outstanding accuracy rate of 96.2%, as shown in Figure 4(a) of the Precision Analysis. On the other hand, ND-ANN achieves a rate of 93.5% in Precision Analysis, surpassing the High-Level Algorithm (HLA), as shown in Figure 4(b). These outcomes demonstrate that the suggested ND-ANN provides more accurate findings, establishing it as a solid and trustworthy technique. Figure 5(a): Accuracy Analysis is compared with ND-ANN Figure 5(b): Accuracy Analysis is compared with HLA In the field of pattern recognition, the accuracy analysis of the Nonlinear Dynamics-Driven Adaptive Neural Network (NDANN) highlights its impressive performance in accurately identifying occurrences across various datasets. An essential measure, accuracy assesses how well the model predicts outcomes on the whole. By analysing its accuracy, one can see that NDANN is quite good at identifying and classifying patterns, as it achieves a high percentage of right classifications. To provide robust learning, generalization, and adaptability to various data features, NDANN's accuracy is enhanced by combining a hybrid learning algorithm with an explainability module, ensemble integration, regularization for stability, and an adaptive architecture. The reliability of NDANN in capturing complicated patterns within real-world data has been confirmed by extensive simulation evaluations, which support its consistent and high accuracy levels across different datasets. Medical diagnosis, financial analytics, and image processing are a few examples of the many fields that can greatly benefit from NDANN's exceptional accuracy in pattern recognition. Based on the results, NDANN is a reliable and accurate tool for many different kinds of applications, and it has the ability to greatly develop pattern recognition technology. Lastly, the accuracy analysis highlights how NDANN is effective at getting accurate and dependable pattern recognition, which shows how it can be used in real-world situations and how it can improve decision-making. The Nonlinear Dynamics- Driven Adaptive Neural Network (ND-ANN) achieved an astounding 98.7 % accuracy, as shown in Figure 5(a) of the Accuracy Analysis. On the other hand, ND-ANN achieves a rate of 93.4% in Accuracy Analysis, surpassing the High-Level Algorithm (HLA), as shown in Figure 5(b). The results show that the proposed ND-ANN is more accurate than other technologies, which proves that it is effective. 0 50 100 150 200 250 300 350 400 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0 A C C U R A C Y A N A L Y S IS R A T IO ( % ) NUMBER OF SAMPLES PR CT-AI FTC RNN ND-ANN 0 50 100 150 200 250 300 350 400 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0 A C C U R A C Y A N A L Y S IS R A T IO ( % ) NUMBER OF SAMPLES PR CT-AI FTC RNN HLA Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 28 https://internationalpubls.com Figure 6(a): Recall Analysis is compared with ND-ANN Figure 6(b): Recall Analysis is compared with HLA In the field of pattern recognition, the recall study of the Nonlinear Dynamics-Driven Adaptive Neural Network (NDANN) reveals its remarkable capacity to discover and retrieve pertinent examples from intricate datasets. An important indicator of performance, recall finds out what percentage of true positives the model got right. NDANN's ability to discover patterns of interest is demonstrated by its recall analysis, which demonstrates its sensitivity to positive instances. To achieve its high recall rates, NDANN combines a hybrid learning algorithm, an explainability module, ensemble integration, regularization for stability, and an adaptive architecture. These features work together to improve the algorithm's ability to identify true positive instances while minimizing the likelihood of false negatives. Robust recall is regularly demonstrated by NDANN through comprehensive simulation evaluations across diverse datasets, showing its ability in reliably capturing a major fraction of positive cases. The memory analysis highlights the usefulness of NDANN in medical diagnostics, anomaly detection, and surveillance systems, among other areas where pattern comprehensiveness is critical. The model is highly effective at identifying relevant instances, making it a helpful tool for applications that require a high level of pattern recognition in complex and ever-changing datasets. Ultimately, the recall analysis confirms that NDANN is capable of accomplishing thorough and trustworthy pattern recognition, highlighting its real-world importance and its capacity to make substantial contributions to improvements in several fields where recall precision is critical. The Nonlinear Dynamics-Driven Adaptive Neural Network (ND- ANN) achieved an impressive recall rate of 96.3%, as shown in Figure 6(a) of the Recall Analysis, highlighting its remarkable performance. On the other hand, ND-ANN achieves a rate of 87.3% in Recall Analysis, surpassing the High-Level Algorithm (HLA), as shown in Figure 6(b). When compared to the current High-Level Algorithm, these results demonstrate how successful the suggested ND-ANN is due to its higher recall capabilities. 0 50 100 150 200 250 300 350 400 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0 R E C A L L A N A L Y S IS R A T IO ( % ) NUMBER OF SAMPLES PR CT-AI FTC RNN ND-ANN 0 50 100 150 200 250 300 350 400 450 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0 R E C A L L A N A L Y S IS R A T IO ( % ) NUMBER OF SAMPLES PR CT-AI FTC RNN HLA Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 29 https://internationalpubls.com Figure 7(a): F1 score Analysis is compared with ND-ANN Figure 7(b): F1 score Analysis is compared with HLA When applied to pattern recognition, the F1 score analysis of the Nonlinear Dynamics-Driven Adaptive Neural Network (NDANN) assesses how well it strikes a balance between the two critical performance metrics of accuracy and recall. By taking precision and recall into account, the F1 score provides a strong indication of the model's capacity to reduce false positives and false negatives at the same time. By analyzing its F1 score, one can see that NDANN is good at balancing recall and accuracy, which allows it to categorize positive cases reliably while minimizing both kinds of classification errors. The impressive F1 scores achieved by NDANN highlight its potential to produce trustworthy pattern recognition results, which are achieved through the utilization of a hybrid learning algorithm, an explainability module, ensemble integration, regularization for stability, and an adaptable architecture. Comprehensive simulation studies on various datasets confirm the model's reliable and impressive F1 score performance, demonstrating its ability to achieve a balanced combination of recall and precision. Because of its capacity to achieve this equilibrium, NDANN is a strong contender for use in security systems, medical diagnosis, and fraud detection, among other areas where preventing either kind of classification error is critical. By providing a comprehensive assessment of its performance and highlighting its practical use in situations where recall and precision are of utmost importance, the F1 score analysis further establishes NDANN as a significant tool for pattern recognition. The Nonlinear Dynamics-Driven Adaptive Neural Network (ND-ANN) achieved an outstanding F1 score of 94.8%, as shown in Figure 7(a) of the F1 Score Analysis. In contrast, Figure 7(b) shows that ND-ANN achieves a score of 90.2% in F1 Score Analysis, surpassing the High-Level Algorithm (HLA). In comparison to the current High-Level Algorithm, these outcomes demonstrate that the suggested ND-ANN provides a better balance between recall and precision, making it a strong and dependable technology. A dependable and strong pattern recognition technology, the Nonlinear Dynamics-Driven Adaptive Neural Network (ND-ANN) shows off great results in F1 score analyses, recall, precision, and accuracy. When it comes to improving pattern recognition technology, ND-ANN is a great tool because of how well it works with different types of datasets and real-world applications like security systems and healthcare diagnostics. In complicated pattern recognition scenarios, the comparative results show that ND-ANN performs better than the High-Level Algorithm (HLA), confirming that it is effective at balancing recall and precision. 0 50 100 150 200 250 300 350 400 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0 F 1 S C O R E A N A L Y S IS R A T IO ( % ) NUMBER OF SAMPLES PR CT-AI FTC RNN ND-ANN 0 50 100 150 200 250 300 350 400 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0 F 1 S C O R E A N A L Y S IS R A T IO ( % ) NUMBER OF SAMPLES PR CT-AI FTC RNN HLA Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 30 https://internationalpubls.com 5. Conclusion Finally, there is a major breakthrough in the area to the study of nonlinear dynamics applied to neural networks for pattern recognition. One new and promising approach is the Nonlinear Dynamics- Driven Adaptive Neural Network (ND-ANN), which is designed to handle complicated and nonlinear patterns found in real-world data. The ND-ANN framework provides a thorough method for using nonlinear dynamics in pattern recognition by solving problems with interpretability, training complexity, and generalization when dealing with noisy data. A complicated and interpretable model was painstakingly sought after by incorporating a hybrid learning method, an explainability module, ensemble integration, regularization for stability, and an adaptive architecture. Extensive simulation analyses showing that ND-ANN is effective across many datasets (in terms of precision, accuracy, recall, and F1 score) validate its superiority over traditional models. The encouraging findings not only show that ND-ANN can properly capture and understand complicated patterns, they set the stage for advancements in pattern recognition technology in the future. Promising advancements in medicine, data processing, and technological capabilities demonstrate the game-changing power of precise nonlinear dynamics modeling. Improved performance and new opportunities for improved applications in many fields are only two reasons why ND-ANN is shining a light on the future of pattern recognition systems. The findings of this study pave the way for future work in the field of advanced and efficient pattern recognition methods by revealing new possibilities for the study and application of nonlinear dynamics in neural networks. References [1] Sui, S., Chen, C. P., & Tong, S. (2020). Neural-network-based adaptive DSC design for switched fractional-order nonlinear systems. IEEE Transactions on Neural Networks and Learning Systems, 32(10), 4703-4712. [2] Zhao, B., Liu, D., & Luo, C. (2019). Reinforcement learning-based optimal stabilization for unknown nonlinear systems subject to inputs with uncertain constraints. IEEE Transactions on Neural Networks and Learning Systems, 31(10), 4330-4340. [3] de Silva, B. M., Champion, K., Quade, M., Loiseau, J. C., Kutz, J. N., & Brunton, S. L. (2020). Pysindy: a python package for the sparse identification of nonlinear dynamics from data. arXiv preprint arXiv:2004.08424. [4] Li, K., Kou, J., & Zhang, W. (2019). Deep neural network for unsteady aerodynamic and aeroelastic modeling across multiple Mach numbers. Nonlinear Dynamics, 96, 2157-2177. [5] Wang, X., Ding, D., Ge, X., & Han, Q. L. (2022). Neural‐network‐based control for discrete‐ time nonlinear systems with denial‐of‐service attack: The adaptive event‐triggered case. International Journal of Robust and Nonlinear Control, 32(5), 2760-2779. [6] Barik D, Saeed MA, Ramachandran T. Experimental and Computational Analysis of Aluminum-Coated Dimple and Plain Tubes in Solar Water Heater System. Energies. 2022 Dec 27;16(1):295. [7] Chen, C., Chen, J., Bao, H., Chen, M., & Bao, B. (2019). Coexisting multi-stable patterns in memristor synapse-coupled Hopfield neural network with two neurons. Nonlinear Dynamics, 95, 3385-3399. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 30 No. 3 (2023) 31 https://internationalpubls.com [8] Li, X., Cao, J., & Ho, D. W. (2019). Impulsive control of nonlinear systems with time- varying delay and applications. IEEE transactions on cybernetics, 50(6), 2661-2673. [9] Arun M, Barik D, Sharma P, GΓΌrel AE, Ağbulut Ü, Medhi BJ, Bora BJ. Experimental and CFD analysis of dimple tube parabolic trough solar water heater with various nanofluids. Applied Nanoscience. 2023 Nov 6:1-47. [10] Wang, M., Wang, Z., Chen, Y., & Sheng, W. (2019). Adaptive neural event-triggered control for discrete-time strict-feedback nonlinear systems. IEEE Transactions on Cybernetics, 50(7), 2946-2958. [11] Wei, T., Li, X., & Stojanovic, V. (2021). Input-to-state stability of impulsive reaction– diffusion neural networks with infinite distributed delays. Nonlinear Dynamics, 103, 1733- 1755. [12] Wang, M., Wang, Z., Chen, Y., & Sheng, W. (2019). Event-based adaptive neural tracking control for discrete-time stochastic nonlinear systems: A triggering threshold compensation strategy. IEEE Transactions on Neural Networks and Learning Systems, 31(6), 1968-1981. [13] Pan, S., & Duraisamy, K. (2020). Physics-informed probabilistic learning of linear embeddings of nonlinear dynamics with guaranteed stability. SIAM Journal on Applied Dynamical Systems, 19(1), 480-509. [14] Arun M, Barik D, Sridhar KP. Experimental and CFD analysis of dimple tube parabolic trough solar collector (PTSC) with TiO2 nanofluids. Journal of Thermal Analysis and Calorimetry. 2022 Dec;147(24):14039-56. [15] Zhao, K., Song, Y., Meng, W., Chen, C. P., & Chen, L. (2020). Low-cost approximation- based adaptive tracking control of output-constrained nonlinear systems. IEEE Transactions on Neural Networks and Learning Systems, 32(11), 4890-4900. [16] Seo, S., Kang, B. S., Lee, J. J., Ryu, H. J., Kim, S., Kim, H., ... & Park, J. H. (2020). Artificial van der Waals hybrid synapse and its application to acoustic pattern recognition. Nature communications, 11(1), 3936. [17] Abiodun, O. I., Jantan, A., Omolara, A. E., Dada, K. V., Umar, A. M., Linus, O. U., ... & Kiru, M. U. (2019). Comprehensive review of artificial neural network applications to pattern recognition. IEEE access, 7, 158820-158846. [18] Quaranta, G., Lacarbonara, W., & Masri, S. F. (2020). A review on computational intelligence for identification of nonlinear dynamical systems. Nonlinear Dynamics, 99(2), 1709-1761. [19] Lin, H., Zhao, B., Liu, D., & Alippi, C. (2020). Data-based fault tolerant control for affine nonlinear systems through particle swarm optimized neural networks. IEEE/CAA Journal of Automatica Sinica, 7(4), 954-964. [20] Salmela, L., Tsipinakis, N., Foi, A., Billet, C., Dudley, J. M., & Genty, G. (2021). Predicting ultrafast nonlinear dynamics in fibre optics with a recurrent neural network. Nature machine intelligence, 3(4), 344-354. [21] Yuan, X., Li, L., & Wang, Y. (2019). Nonlinear dynamic soft sensor modeling with supervised long short-term memory network. IEEE transactions on industrial informatics, 16(5), 3168-3176.