Weights Of Neural Network Research Articles

A Continuous Remaining Useful Life Prediction Method With Multistage Attention Convolutional Neural Network and Knowledge Weight Constraint.

The rotating machinery is continuously monitored in practical application. However, the historical life-cycle data cannot be always preserved due to the limited storage resource; meanwhile, the on-site computing platform cannot process a large number of monitoring samples. It brings a great challenge for the remaining useful life (RUL) prediction. Thus, continuous learning (CL) is introduced into RUL prediction model for achieving its knowledge accumulation and dynamic update. To improve the performance of continuous RUL prediction, this article presents a new RUL prediction methodology with a multistage attention convolutional neural network (MSACNN) and knowledge weight constraint (KWC). First, an improved multihead full-channel sight self-attention (MFCSSA) mechanism is proposed to capture the global degradation information across all channels. MSACNN is then constructed by embedding MFCSSA, squeeze-and-excitation (SE) mechanism, and convolutional block attention module (CBAM) into different stages of feature extraction, which enables it to capture the global degradation information and refine the feature representations progressively. The KWC mechanism based on the importance of weight parameters and gradient information is proposed and integrated into MSACNN to achieve the continuous RUL prediction task. The proposed KWC can effectively alleviate catastrophic forgetting in CL. Finally, the experimental results on the life-cycle bearing and gear datasets demonstrate that MSACNN has a higher accuracy than the existing prediction methods. Moreover, the KWC mechanism performs better than typical CL methods in retaining the previously learned knowledge while acquiring the new task knowledge. Therefore, the proposed methodology can be better applied to the continuous RUL prediction tasks than the advanced methods of the same kind.

Decentralized Event-Triggered Tracking Control for Unmatched Interconnected Systems via Particle Swarm Optimization-Based Adaptive Dynamic Programming.

The problem of the large-scale interconnected system (LSIS) control is prevalent in practical engineering and is becoming increasingly complex. In this article, we propose a novel decentralized event-triggered tracking control (ETTC) strategy for a class of continuous-time nonlinear LSIS with unmatched interconnected terms and asymmetric input constraints. First, auxiliary subsystems are established to address the unmatched cross-linking terms. Next, the dynamics states of the tracking error and the exosystem are combined to construct a nominal augmented subsystem. By employing a nonquadratic performance function, the input-constrained decentralized tracking control problem is transformed into an optimal control problem for the nominal augmented subsystem. A group of independent parameters and event-triggered conditions are designed to save communication bandwidth and computational resources. Subsequently, the critic-only adaptive dynamic programming (ADP) method is used to solve the Hamilton-Jacobi-Bellman equation (HJBE) associated with the optimal control problem. To improve training success rate, the weights of the critic neural network (NN) are updated by introducing a particle swarm optimization algorithm (PSOA). The tracking error and the NN weights are proved to be uniformly ultimately bounded (UUB) under the proposed ETTC by using the Lyapunov extension theorem. Finally, the simulation example of an unmatched interconnected system is provided to verify the validity of the proposed decentralized method.

Reduced-Order Neural Network Synthesis With Robustness Guarantees.

In the wake of the explosive growth in smartphones and cyber-physical systems, there has been an accelerating shift in how data are generated away from centralized data toward on-device-generated data. In response, machine learning algorithms are being adapted to run locally on board, potentially hardware-limited, devices to improve user privacy, reduce latency, and be more energy efficient. However, our understanding of how these device-orientated algorithms behave and should be trained is still fairly limited. To address this issue, a method to automatically synthesize reduced-order neural networks (having fewer neurons) approximating the input-output mapping of a larger one is introduced. The reduced-order neural network's weights and biases are generated from a convex semidefinite program that minimizes the worst case approximation error with respect to the larger network. Worst case bounds for this approximation error are obtained and the approach can be applied to a wide variety of neural networks architectures. What differentiates the proposed approach to existing methods for generating small neural networks, e.g., pruning, is the inclusion of the worst case approximation error directly within the training cost function, which should add robustness to out-of-sample data points. Numerical examples highlight the potential of the proposed approach. The overriding goal of this article is to generalize recent results in the robustness analysis of neural networks to a robust synthesis problem for their weights and biases.

SSA‐BP Neural Network Model for Predicting Rice‐Fish Production in China

The rice‐fish system has gained significant interest in recent years because of its effective usage of limited land and freshwater resources. To scientifically guide the improvement of rice field fishery production, the data in this study were selected from the latest China Fishery Statistical Yearbook, and therefore the development of paddy aquaculture was investigated. In order to more precisely predict the production of rice‐fish in China, this paper introduces an artificial neural network with the SSA‐BP model, which solves the drawbacks of the BP neural network such as easy to fall into local optimum and slow convergence speed when it is used for prediction. Firstly, the SSA‐BP model incorporates the aquaculture area (split by water area), the national freshwater fish seedling output, the national end‐of‐year ownership of inland fishing vessels, the number of freshwater fisheries practitioners as input variables, and the production of rice‐fish as an output variable; secondly, the SSA optimization algorithm was used to find the optimal initial thresholds and weights for the BP neural network, and finally the SSA‐BP prediction model was constructed. The results revealed that the overall expansion of the rice field fishery was swift in the last five years, and the output of cultivated fish in China’s rice fields rose by nearly 20% yearly in the past five years. Compared with the BP neural network and GA‐BP models, the accuracy of the SSA‐BP prediction model was enhanced by 61.01% and 16.15%, respectively, which was more suited for predicting the production of rice‐fish.

PHYSICS-INFORMED DEEP AI SIMULATION FOR FRACTAL INTEGRO-DIFFERENTIAL EQUATION

Fractal integro-differential equations (IDEs) can describe the effect of local microstructure on a complex physical problem, however, the traditional numerical methods are not suitable for solving the new-born models with the fractal integral and fractal derivative. Here we show that deep learning can be used to solve the bottleneck. By the two-scale transformation, the fractal IDE is first approximately converted to its traditional integro-differential partner, which is further converted to a differential equation system by introducing an auxiliary variable to remove the integral operation. Moreover, a flexible adaptive technology is adopted to deal with the loss weights of a deep learning neural network. A fractal Volterra IDE is used to show the effectiveness and simplicity of this new physics-informed deep AI simulation model. All results indicate the AI simulation model has good robustness and convergence, and the fractal Volterra IDE might explore the different properties of viscoelasticity for a porous medium.

Event-Based Robust Optimal Consensus Control for Nonlinear Multiagent System With Local Adaptive Dynamic Programming.

This article investigates the robust optimal consensus for nonlinear multiagent systems (MASs) through the local adaptive dynamic programming (ADP) approach and the event-triggered control method. Due to the nonlinearities in dynamics, the first part defines a novel measurement error to construct a distributed integral sliding-mode controller, and the consensus errors can approximately converge to the origin in a fixed time. Then, a modified cost function with augmented control is proposed to deal with the unmatched disturbances for the event-based optimal consensus controller. Specifically, a single network local ADP structure with novel concurrent learning is presented to approximate the optimal consensus policies, which guarantees the robustness of the MASs and the uniform ultimate boundedness (UUB) of the neural network (NN) weights' estimation error and relaxes the requirement of initial admissible control. Finally, an illustrative simulation verifies the effectiveness of the method.

Exploring the effects of diverse historical stock price data on the accuracy of stock price prediction models

In the recent years, a majority of Americans have moved to investing in stocks through algorithmic trading. Not only does this underline the public’s interest in the stock market, but it also shows the important role that artificial intelligence plays when it comes to predicting and trading stocks. These prediction models can benefit investors by improving their investment decisions and potentially increasing financial gains. In this research, we aimed to explore the impact of incorporating historical stock price data types, such as the opening prices, closing prices and highest prices, on the accuracy of stock price prediction models. The primary objective was to optimize the performance of our models. To carry out this research, we developed three supervised learning artificial intelligence models using Python: linear regression, neural network, and multiplicative weight update. Our models predicted the stock prices for Microsoft, Amazon, Google, Tesla, and Apple. First, the models used the opening prices for the past three days to predict the opening price on the fourth day. However, to enhance each model’s performance, we evaluated whether adding extra information, such as closing and highest prices, would be beneficial. We hypothesized that incorporating the opening, closing, and highest prices would yield the highest accuracy as it would provide the models with the most information and help them better predict patterns in the stock market. The results supported our hypothesis as the models’ average percent errors significantly decreased when they were given all three of these data types.

Base on -Learning Pareto Optimality for Linear Itô Stochastic Systems With Markovian Jumps

This article investigate the cooperative differential game (CDG) for continuous-time linear Itô stochastic systems with markovian jumps (SSMJ) to obtain the Pareto solutions. Different from most existing works studying nonzero-sum games, this article studies the CDG on the quadratic infinite horizon for the Itô-type SSMJ with unknown system matrix and transition probability. A novel <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$Q$</tex-math> </inline-formula> -learning online algorithm is developed, which consists of that (i) the optimal control problem is equivalent to solving a stochastic algebraic Riccatic equation (ARE); (ii) the joint cost function is approximated by a critic neural network (NN) and Pareto efficient is approximated by two actor NNs. The rigorous stability analysis shows that the system state for SSMJ and the NN weight errors are uniformly ultimately bounded (UUB). Finally, the theory analysis is validated by a numerical example with detailed discussions. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —In practical applications, many systems are often affected by the change of external environment or the failure of internal components, which leads to the random jump of system parameters. Markovian jump system can effectively describe the above problems. And when the system model is disturbed by the internal parameters, the system control input and external environment, the random errors of state measurement, and other random factors, the deterministic model can no longer accurately describe the controlled system. Therefore, the SSMJ can describe practical problems more accurately. By cooperation, in general, the cost one specific player incurs is not uniquely determined anymore. If all players decide, for example, to use their control variables to reduce the cost of player 1 as much as possible, a different minimum is attained for player 1 compared with that in the case where all players agree collectively to help a different player in minimizing his cost. So, depending on how the players choose to ‘divide’ their control efforts, a player incurs different ‘minima’. Therefore, we will design an online learning algorithm to obtain pareto solutions with different weights. On the other hand, in practice, it is difficult to obtain an accurate system model. In order to solve this problem, a novel scheme is designed by using <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$Q$</tex-math> </inline-formula> -learning technology, which does not need system matrix.

Robust Adaptive Leader-Following Formation Control of Nonlinear Multiagents Using Three-Layer Neural Networks.

This article studies a formation control problem for a group of heterogeneous, nonlinear, uncertain, input-affine, second-order agents modeled by a directed graph. A tunable neural network (NN) is presented, with three layers (input, two hidden, and output) that can approximate an unknown nonlinearity. Unlike one- or two-layer NNs, this design has the advantage of being able to set the number of neurons in each layer ahead of time rather than relying on trial and error. The NN weights tuning law is rigorously derived using the Lyapunov theory. The formation control problem is tackled using a robust integral of the sign of the error feedback and NNs-based control. The robust integral of the sign of the error feedback compensates for the unknown dynamics of the leader and disturbances in the agent errors, while the NN-based controller accounts for the unknown nonlinearity in the multiagent system. The stability and semi-global asymptotic tracking of the results are proven using the Lyapunov stability theory. The study compares its results with two others to assess the effectiveness and efficiency of the proposed method.

Reinforcement Learning Control for a 2-DOF Helicopter With State Constraints: Theory and Experiments

This study focuses on the novel reinforcement learning control strategy of a nonlinear two-degrees-of-freedom (2-DOF) helicopter system for tracking the desired trajectory while minimizing the tracking error. First, gradient descent algorithm is incorporated in the context of the reinforcement learning control scheme to obtain the adaptive laws. Subsequently, considering the uncertainties in the nonlinear system, radial basis function (RBF) neural networks (NNs) are exploited to approximate the unknown internal dynamics. In contrast to the previous studies, aiming at accelerating the convergence in reinforcement learning control, a barrier Lyapunov function is constructed to constrain the states to ensure that the tracking error rapidly converges to a neighborhood of zero. Under the proposed control strategy, the states of the closed-loop system are proven to be semi-globally uniformly ultimately bounded through rigorous Lyapunov analyses, and the state constraints are satisfied. Furthermore, the simulations and experiments conducted on a Quanser laboratory platform reveal that the proposed control functions are suitable and effective. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —This paper is motivated by designing a reinforcement learning control strategy to enhance online learning capability and control performance of the controller for a nonlinear 2-DOF helicopter system. The control framework is divided into the design of the critic and actor NNs, responsible primarily for evaluating the control performance and approximating uncertainties in the system separately. Unlike the adaptive NN control, the actor NN weights are updated by combining information of states and inputs from the critic NN. In addition, aiming at accelerating the convergence, a barrier Lyapunov function is constructed to constrain the states to ensure that the tracking error rapidly converges to a neighborhood of zero. Finally, the proposed control strategy is validated in simulation and experiment on the Quanser laboratory platform.

A study of key issues in parallel algorithms for face recognition based on genetic neural networks

Abstract This study examines the effectiveness of Genetic Neural Networks (GNN) in face recognition, particularly in optimizing parallel algorithms to overcome the challenges posed by complex data. We have significantly improved recognition accuracy and computational efficiency by employing an adaptive genetic algorithm that fine-tunes neural network weights through Selection, crossover, and mutation. Our approach was tested across diverse datasets, covering variations in posture, age, ethnicity, and lighting conditions. The results demonstrate outstanding recognition rates: 99.82% on LFW, 97.94% on AgeDB-30, 95.11% on CFP-FP, 95.87% on CALFW, and 89.44% on CPLFW, showcasing exceptional robustness against complex lighting and occlusions. Additionally, our algorithm maintains balanced accuracy across different ethnicities with an overall recognition rate of 96.77% and boasts a substantial reduction in processing time to an average of 4.15 seconds. These advancements underscore the potential and practicality of our method in enhancing face recognition technology.

Application of QGA-BP Neural Network in Debt Risk Assessment of Government Platforms

How to correctly understand the existence of local government debt, study its risk classification and impact, give full play to the “dual nature” of debt with a full-caliber indicator system, and avoid debt risks to the greatest extent. That is the research direction of this article. In order to improve the accuracy and efficiency of risk assessment and effectively reduce the debt risk of government platform companies, a risk assessment method based on optimized back-propagation (BP) neural network is proposed. First, the method uses quantum genetic algorithm (quantum genetic algorithm, QGA) to adjust and determine the initial weight and threshold of BP neural network and realize the optimization of BP neural network model parameter setting. Then, the QGA-BP debt risk assessment of government platforms is verified that it performs well in the debt risk prediction of government platform companies, and its prediction accuracy and prediction speed are improved.

An efficient model‐free adaptive optimal control of continuous‐time nonlinear non‐zero‐sum games based on integral reinforcement learning with exploration

AbstractTo reduce the learning time and space occupation, this study presents a novel model‐free algorithm for obtaining the Nash equilibrium solution of continuous‐time nonlinear non‐zero‐sum games. Based on the integral reinforcement learning method, a new integral HJ equation that can quickly and cooperatively determine the Nash equilibrium strategies of all players is proposed. By leveraging the neural network approximation and gradient descent method, simultaneous continuous‐time adaptive tuning laws are provided for both critic and actor neural network weights. These laws facilitate the estimation of the optimal value function and optimal policy without requiring knowledge or identification of the system's dynamics. The closed‐loop system stability and convergence of weights are guaranteed through the Lyapunov analysis. Additionally, the algorithm is enhanced to reduce the number of auxiliary NNs used in the critic. The simulation results for a two‐player non‐zero‐sum game validate the effectiveness of the proposed algorithm.

A Model of Continual and Deep Learning for Aspect Based in Sentiment Analysis

Sentiment Analysis is a useful tool in several social and business contexts. Aspect Sentiment Classification is a subtask in Sentiment Analysis that gives information about features or aspects of people, entities, products, or services present in reviews. Different Deep Learning models have been proposed to solve Aspect Sentiment Classification focus on a specific domain such as restaurant,hotel, or laptop reviews. However, there are few proposals for creating a single model with high performance in multiple domains. The Continual Learning approach with neural networks has been used to solve aspect classification in multiple domains. However, avoid low aspect classification performance in Continual Learning is challenging. As a consequence, potential neural networkweight shifts in the learning process in different domains or datasets.In this paper, a novel Aspect Sentiment Classification approach is proposed. Our approach combines a Transformer Deep Learning technique with a Continual Learning algorithm in different domains. The input layer used is the pre‐trained model Bidirectional Encoder Representations from Transformers. The experiments show the efficacy of our proposal with 78 .

A Model of Continual and Deep Learning for Aspect Based in Sentiment Analysis

Sentiment Analysis is a useful tool in several social and business contexts. Aspect Sentiment Classification is a subtask in Sentiment Analysis that gives information about features or aspects of people, entities, products, or services present in reviews. Different Deep Learning models have been proposed to solve Aspect Sentiment Classification focus on a specific domain such as restaurant,hotel, or laptop reviews. However, there are few proposals for creating a single model with high performance in multiple domains. The Continual Learning approach with neural networks has been used to solve aspect classification in multiple domains. However, avoid low aspect classification performance in Continual Learning is challenging. As a consequence, potential neural networkweight shifts in the learning process in different domains or datasets.In this paper, a novel Aspect Sentiment Classification approach is proposed. Our approach combines a Transformer Deep Learning technique with a Continual Learning algorithm in different domains. The input layer used is the pre‐trained model Bidirectional Encoder Representations from Transformers. The experiments show the efficacy of our proposal with 78 .

Attention network for predicting T-cell receptor-peptide binding can associate attention with interpretable protein structural properties.

Understanding how a T-cell receptor (TCR) recognizes its specific ligand peptide is crucial for gaining an insight into biological functions and disease mechanisms. Despite its importance, experimentally determining TCR-peptide-major histocompatibility complex (TCR-pMHC) interactions is expensive and time-consuming. To address this challenge, computational methods have been proposed, but they are typically evaluated by internal retrospective validation only, and few researchers have incorporated and tested an attention layer from language models into structural information. Therefore, in this study, we developed a machine learning model based on a modified version of Transformer, a source-target attention neural network, to predict the TCR-pMHC interaction solely from the amino acid sequences of the TCR complementarity-determining region (CDR) 3 and the peptide. This model achieved competitive performance on a benchmark dataset of the TCR-pMHC interaction, as well as on a truly new external dataset. Additionally, by analyzing the results of binding predictions, we associated the neural network weights with protein structural properties. By classifying the residues into large- and small-attention groups, we identified statistically significant properties associated with the largely attended residues such as hydrogen bonds within CDR3. The dataset that we created and the ability of our model to provide an interpretable prediction of TCR-peptide binding should increase our knowledge about molecular recognition and pave the way for designing new therapeutics.

The Capabilities of Boltzmann Machines to Detect and Reconstruct Ising System's Configurations from a Given Temperature.

The restricted Boltzmann machine (RBM) is a generative neural network that can learn in an unsupervised way. This machine has been proven to help understand complex systems, using its ability to generate samples of the system with the same observed distribution. In this work, an Ising system is simulated, creating configurations via Monte Carlo sampling and then using them to train RBMs at different temperatures. Then, 1. the ability of the machine to reconstruct system configurations and 2. its ability to be used as a detector of configurations at specific temperatures are evaluated. The results indicate that the RBM reconstructs configurations following a distribution similar to the original one, but only when the system is in a disordered phase. In an ordered phase, the RBM faces levels of irreproducibility of the configurations in the presence of bimodality, even when the physical observables agree with the theoretical ones. On the other hand, independent of the phase of the system, the information embodied in the neural network weights is sufficient to discriminate whether the configurations come from a given temperature well. The learned representations of the RBM can discriminate system configurations at different temperatures, promising interesting applications in real systems that could help recognize crossover phenomena.

Reinforcement learning-based formation-surrounding control for multiple quadrotor UAVs pursuit-evasion games

This paper proposes a reinforcement learning-based formation-surrounding control method for multiple quadrotor unmanned aerial vehicles (UAVs) pursuit-evasion (MPE) games system subject to external disturbances. In the framework of the MPE games, the pursuers aim to equally surround the evaders which try to avoid being surrounded when forming the desired formation. By constructing position and attitude tracking error subsystems of quadrotor UAV, this paper proposes two control strategies which combines the feedforward control technique and reinforcement learning (RL) method. First, two novel cost functions are presented for the quadrotor UAV with external disturbances. Then, two control schemes based on RL have been developed to guarantee the stability of the tracking error subsystem. Subsequently, two critic-only neural networks (NN) weight update laws that only satisfy finite excitation conditions are proposed to estimate the optimal cost function. Furthermore, Nash equilibrium for multiple quadrotor UAVs is achieved by means of RL strategy to solve the Hamilton-Jacobi-Isaacs (HJI) equations. And the property of equally surrounding is proved for the first time by utilizing Euler's formula in this paper. Finally, the numerical simulation results are given to show the effectiveness and superior performance of the proposed control method.

Visual Servoing Control of Robotics With A Neural Network Estimator Based on Spectral Adaptive Law

In recent years, with the development of machine vision and other relative techniques, visual servoing control of robotics has been wildly applied. A complex calibration process is usually required to get the accurate parameters of the camera and the robot, so that getting the projection relationship between changes of images and movement of robot joints usually takes much effort. In order to solve this problem, a ReLU activating neural network (NN) estimator is proposed to estimate the compound Jacobian matrix of the system in this paper. The weight of the NN is updated online by a project algorithm with a novel spectral adaptive law which can effectively improve the generalization ability of the NN and the robustness of the system. By constructing a new Lyapunov function with the spectral norm of weight of NN, the stability of the proposed adaptive algorithm and the controller can be proved. Simulations and experimental results validate the effectiveness of the proposed controller.

Multi-source underwater DOA estimation using PSO-BP neural network based on high-order cumulant optimization

Due to the complex and changeable environment under water, the performance of traditional DOA estimation algorithms based on mathematical model, such as MUSIC, ESPRIT, etc., degrades greatly or even some mistakes can be made because of the mismatch between algorithm model and actual environment model. In addition, the neural network has the ability of generalization and mapping, it can consider the noise, transmission channel inconsistency and other factors of the objective environment. Therefore, this paper utilizes Back Propagation (BP) neural network as the basic framework of underwater DOA estimation. Furthermore, in order to improve the performance of DOA estimation of BP neural network, the following three improvements are proposed. (1) Aiming at the problem that the weight and threshold of traditional BP neural network converge slowly and easily fall into the local optimal value in the iterative process, PSO-BP-NN based on optimized particle swarm optimization (PSO) algorithm is proposed. (2) The Higher-order cumulant of the received signal is utilized to establish the training model. (3) A BP neural network training method for arbitrary number of sources is proposed. Finally, the effectiveness of the proposed algorithm is proved by comparing with the state-of-the-art algorithms and MUSIC algorithm.