Weight Update Algorithm Research Articles

Ferrimagnet-Based Neuromorphic Device Mimicking the Ventral Visual Pathway for High-Accuracy Target Recognition.

The ventral visual pathway (VVP) of the human brain efficiently implements target recognition by employing a deep hierarchical structure to build complex visual concepts from simple features. Artificial neural networks (ANNs) based on spintronic devices are capable of target recognition, but their poor interpretability and limited network depth hinder ANNs from mimicking the VVP. Hardware implementation of the VVP requires a biorealistic spintronic device as well as the corresponding interpretable and deep network structure, which have not been reported so far. Here, we report a ferrimagnetic neuron with a continuously differentiable exponential linear unit (CeLu) activation function, which is closer to biological neurons and could mitigate the issue of limited network depth. Meanwhile, we also demonstrate that a ferrimagnet can construct artificial synapses with high linearity and symmetry to meet the requirements of weight update algorithms. Based on these neurons and synapses, we propose an all-spin convolutional neural network (CNN) with a high interpretability and deep neural network, to mimic the VVP. Compared to the state-of-the-art spintronic-based neuromorphic computing model, the CNN with bionic function, using experimentally derived device parameters, achieves high recognition accuracies of over 91% and 98% on the CIFAR-10 datasets and the MNIST datasets, respectively, showing improvements of 1.13% and 1.76%. Our work provides a promising method to improve the bionic performance of spintronic device-based neural networks.

A spiking neural network with continuous local learning for robust online brain machine interface

Objective. Spiking neural networks (SNNs) are powerful tools that are well suited for brain machine interfaces (BMI) due to their similarity to biological neural systems and computational efficiency. They have shown comparable accuracy to state-of-the-art methods, but current training methods require large amounts of memory, and they cannot be trained on a continuous input stream without pausing periodically to perform backpropagation. An ideal BMI should be capable training continuously without interruption to minimize disruption to the user and adapt to changing neural environments. Approach. We propose a continuous SNN weight update algorithm that can be trained to perform regression learning with no need for storing past spiking events in memory. As a result, the amount of memory needed for training is constant regardless of the input duration. We evaluate the accuracy of the network on recordings of neural data taken from the premotor cortex of a primate performing reaching tasks. Additionally, we evaluate the SNN in a simulated closed loop environment and observe its ability to adapt to sudden changes in the input neural structure. Main results. The continuous learning SNN achieves the same peak correlation () as existing SNN training methods when trained offline on real neural data while reducing the total memory usage by 92%. Additionally, it matches state-of-the-art accuracy in a closed loop environment, demonstrates adaptability when subjected to multiple types of neural input disruptions, and is capable of being trained online without any prior offline training. Significance. This work presents a neural decoding algorithm that can be trained rapidly in a closed loop setting. The algorithm increases the speed of acclimating a new user to the system and also can adapt to sudden changes in neural behavior with minimal disruption to the user.

Adaptive Feed-Forward Control for Gust Load Alleviation on a Flying-Wing Model Using Multiple Control Surfaces

Based on measured gust information, a multi-input multi-output (MIMO) adaptive feed-forward control scheme for gust load alleviation (GLA) on a semi-span flying-wing aircraft using multiple control surfaces is proposed. In order to remedy weight drift and biased estimation problems that are commonly encountered in adaptive control, the circular leaky LMS (CLLMS) algorithm is employed, which utilizes gust measurement information, filtered reference signals, and error signals to update controller parameters online. The results demonstrate that good load reductions are achieved in both continuous and discrete gust environments. For instance, the designed GLA control system leads to an 80.72% reduction in the root-mean-square (RMS) values of wing-root bending moment in the Dryden gust environment and a 77.59% reduction of its maximum value in the 1-cos discrete gust condition. Based on the limited power of the actuator and the limited authority for control surface deflections when integrating GLA into the flight control system, a weight-updating algorithm with deflection angle and rate constraints on control surfaces is proposed. The simulation results show that the strict constraints on control surface deflections will degrade the GLA performance. Finally, the influence of the partial jamming fault of actuators on GLA performance is studied. It is found that good GLA performance can be preserved despite the degraded performance during the initial stage of the actuator jamming fault. This is due to the robustness brought about by multiple control surfaces and the adaptability of the control algorithm.

Neural Network Adaptive Observer design for Nonlinear Systems with Partially and Completely Unknown Dynamics Subject to Variable Sampled and Delay Output Measurement

This paper proposes a novel Neural Network Adaptive Observer (NNAO) for Nonlinear Systems with Partially and Completely Unknown Dynamics (NSPCUD), subject to variable sampled and delayed output. The method involves designing a neural network observer for partially unknown nonlinear systems with sampled and delayed outputs, using a radial basis function (RBF) neural network to approximate the system’s unknown part. A new weight update algorithm is proposed, along with a closed-loop output predictor for coping with variable samples, and a closed-loop integral compensation to handle variable delay. This approach is then extended to cover completely unknown systems as well. Numerical simulations and comparisons between the proposed method and previous methods on autonomous ground vehicle models were conducted to verify the effectiveness of the proposed NNAO.

Adaptive echo state network with a recursive inverse-free weight update algorithm

Echo State Network (ESN) is widely applied in sequence prediction, physics, and economics. Moore-Penrose inversion is a typical solving process of ESN. However, in the era of big data, the traditional inversion is computationally costly and the data may be irreversible. To solve this problem, a recursive inverse-free algorithm ESN model, called RIFESN, is introduced in this article. The model also introduces an adaptive activation function based on the error between target value and predicted value, which greatly improves the nonlinear mapping capability of the network. Three simulation experiments are conducted on the Mackey-Glass-17 time series, sunspots series and daily exchange rate summary of eight countries from 1990 to 2016, respectively. The results of the three simulations show that RIFESN has better prediction accuracy and operational speed, which proves that it has a good application prospect.

A novel neural emulator identification of nonlinear dynamical systems using Lyapunov stability theory

This paper deals with a new weight-updating algorithm using Lyapunov stability theory (LST) for the training of a neural emulator (NE), of nonlinear systems, connected by an autonomous algorithm inspired from the real-time recurrent learning (RTRL). The proposed method is formulated by an inequality-constraint optimization problem where the Lagrange multiplier theory is used as the optimization tool. The contribution of this paper is the integration of the LST into the Lagrange constraint function to synthesize a new analytical adaptation gain rate satisfying the asymptotic stability of the NE and providing good emulation performances. To confirm the good performances and the convergence ability of the proposed adaptation algorithm, a numerical example and an experimental validation on a chemical reactor are proposed.

Parallel Network-Based Sliding Mode Tracking Control for Robotic Manipulators with Uncertain Dynamics

Robot dynamics model uncertainty and unpredictable external perturbations are important factors that influence control accuracy and stability. To accurately compensate for the dynamics model in sliding mode control (SMC), a new parallel network (PCR) is proposed in this paper. The network parallelizes the radial basis function and convolutional neural network, which gives it the advantage of making full use of one-dimensional data fitting results and two-dimensional data feature information, realizing the deep learning of multidimensional data and improving the model’s compensation accuracy and anti-interference ability. Meanwhile, based on the integration of adaptive control techniques and gradient descent, a new weight update algorithm is designed to realize the online learning of PCR networks under loss-free functions. Then, a new sliding mode controller (PCR-SMC) is established. The model-free intelligent control of the robot is accomplished without knowledge of the predetermined upper bounds. Additionally, the stability analysis of the control system is proved by the Lyapunov theorem. Lastly, robot tracking control simulations are performed on two trajectories. The results demonstrate the high-precision tracking performance of this controller in comparison with the RBF-SMC controller.

LNS-Madam: Low-Precision Training in Logarithmic Number System Using Multiplicative Weight Update

Representing deep neural networks (DNNs) in low-precision is a promising approach to enable efficient acceleration and memory reduction. Previous methods that train DNNs in low-precision typically keep a copy of weights in high-precision during the weight updates. Directly training with low-precision weights leads to accuracy degradation due to complex interactions between the low-precision number systems and the learning algorithms. To address this issue, we develop a co-designed low-precision training framework, termed LNS-Madam, in which we jointly design a logarithmic number system (LNS) and a multiplicative weight update algorithm (Madam). We prove that LNS-Madam results in low quantization error during weight updates, leading to stable performance even if the precision is limited. We further propose a hardware design of LNS-Madam that resolves practical challenges in implementing an efficient datapath for LNS computations. Our implementation effectively reduces energy overhead incurred by LNS-to-integer conversion and partial sum accumulation. Experimental results show that LNS-Madam achieves comparable accuracy to full-precision counterparts with only 8 bits on popular computer vision and natural language tasks. Compared to FP32 and FP8, LNS-Madam reduces the energy consumption by over 90% and 55%, respectively.

Privacy information protection algorithm of ultra dense network nodes based on edge computing

In order to solve the problems of high node loss rate, high time overhead and high risk of privacy disclosure in network node privacy information protection, an ultra dense network node privacy information protection algorithm based on edge computing is proposed. The weight update algorithm is used to detect the security vulnerabilities of ultra dense network nodes. According to the detection results, the characteristics of node vulnerabilities are obtained, the sensitive label information of node vulnerabilities is protected through the weighted graph, and the k-anonymity technology is used to anonymize the privacy information of ultra dense network nodes; Finally, edge computing is used to protect the privacy information of nodes. The experimental results show that the node loss rate of the proposed method is always less than 2%, the time overhead is small, and the risk coefficient of privacy disclosure is small.

Self-Triggered Adaptive NN Tracking Control for a Class of Continuous-Time Nonlinear Systems With Input Constraints

This article develops a self-triggered adaptive neural network (NN) tracking controller for a class of continuous-time nonlinear systems, that is, input constrained and with unknown drift and input dynamics. Since the drift and input dynamics are both unknown, an NN is built within a self-triggered update paradigm to approximate the unknown tracking control. The error derivative used in the weight update algorithm is derived using a robust exact differentiator technique. To address input constraints, an auxiliary compensator is designed for the unimplemented control effort. Through rigorous Lyapunov analyses, we can guarantee that all the tracking and weight errors are uniformly ultimately bounded. Finally, to show the effectiveness of the proposed control performance, simulation results of a two-link robot are provided and analyzed.

A Spike Neural Network Model for Lateral Suppression of Spike-Timing-Dependent Plasticity with Adaptive Threshold

Aiming at the practical constraints of high resource occupancy and complex calculations in the existing Spike Neural Network (SNN) image classification model, in order to seek a more lightweight and efficient machine vision solution, this paper proposes an adaptive threshold Spike Neural Network (SNN) model of lateral inhibition of Spike-Timing-Dependent Plasticity (STDP). The conversion from grayscale image to pulse sequence is completed by convolution normalization and first pulse time coding. The network self-classification is realized by combining the classical Spike-Timing-Dependent Plasticity algorithm (STDP) and lateral suppression algorithm. The occurrence of overfitting is effectively suppressed by introducing an adaptive threshold. The experimental results on the MNIST data set show that compared with the traditional SNN classification model, the complexity of the weight update algorithm is reduced from O(n2) to O(1), and the accuracy rate can still remain stable at about 96%. The provided model is conducive to the migration of software algorithms to the bottom layer of the hardware platform, and can provide a reference for the realization of edge computing solutions for small intelligent hardware terminals with high efficiency and low power consumption.

Development of Adaptive Testing Method Based on Neurotechnologies

Purpose of the study. The aim of the study is to create neural network models of modules in an adaptive testing system to design an individual testing trajectory.The research article discusses the implementation of an adaptive testing system in terms of introducing artificial neural network modules into its composition, which should solve the problem of choosing a topic and the complexity of the next question, taking into account previous answers and the complexity of previously asked questions, as well as the connectivity of topics and response time as a factor guessing or searching for an answer, thereby forming an individual testing trajectory.Materials and methods. In the course of the study, the data that affect the quality of the solution of the problem was analyzed, the general modular structure of the system was proposed, and the main data flows entering the input of an artificial neural network (ANN) were described. To solve the problem of choosing the complexity of a question, it is proposed to use a feed-forward network, a comparison of various ANN architectures and training parameters (weight update algorithms, loss functions, number of training epochs, packet sizes) is carried out. As an alternative, the possibility of using a recurrent ANN LSTM (Long-Short Term Memory) network is considered. All results were obtained using the high-level Keras library, which allows you to quickly start at the initial stages of research and get the first results. SGD, Adam, NAdam and RMSprop implemented in Keras were compared as optimizers to achieve faster convergence. Adam showed the best results in terms of accuracy, while the MSE loss function (mean square error) was used together with the optimizer. Traditionally, training was carried out for a large number of epochs; graphs of dependences of accuracy on the number of epochs for a different number of neurons in the hidden layer were experimentally obtained.Results. Based on the study, we can conclude that the obtained accuracy of the direct propagation network of 80-85% is quite sufficient for its use in the adaptive testing system. However, it remains to answer the question of the need to improve the efficiency of an already implemented network, and, therefore, to conduct research on methods to improve the efficiency of networks, including finer tuning of parameters and learning algorithms, as well as architecture.A well-known and obvious drawback of using LSTMs is their exactingness in terms of equipment and resources, both during training (the training process takes a significant amount of time) and during startup, in our case, it is supplemented by increased requirements for the training sample and casts doubt on the advisability of further study of LSTM networks when solving this task.Conclusion. The introduction of the proposed tools will allow implementing an adaptive testing system, with an intelligent selection of questions depending on the demonstrated level of knowledge of the test person to form an individual testing trajectory in order to determine the reliable level of knowledge of the test subject for the optimal number of questions asked.

Intelligent optimal hybrid motion/force control of constrained robot manipulator

This paper presents a neural network-based optimal control approach for hybrid motion/force control problem of robot manipulators in the presence of structured and unstructured uncertainties. The quadratic optimisation with sliding mode control and neural network is utilised to propose an optimal hybrid motion/force control of robotic manipulators. Firstly, the dynamics of robot manipulator is reduced into state-space form describing the constrained and unconstrained motion separately. Then the optimal hybrid motion/force control scheme is derived utilising the optimisation of Hamilton Jacobi Bellman (HJB) equation and sliding mode control approach. The structured and unstructured uncertainties of the system are compensated using radial basis function neural network and adaptive compensator. The radial basis function neural network approximates the unknown dynamics and adaptive compensator is used to estimate the bounds on neural network approximation error and the unstructured uncertainties of the system. The neural networks are trained in online manner using weight update algorithms derived with Lyapunov approach to guarantee the tracking stability and error convergence with prescribed quadratic performance index. Finally, the proposed control approach is verified through simulation results with two link constrained manipulator in a comparative manner.

The Study on Void Fraction Prediction of Gas-liquid Two Phase Flow Based on Convolutional Neural Network

For the gas-liquid two phase flow in the horizontal pipeline, at the center angle the void fraction of the different liquid phases is calculated with the finite element simulation software, and then a soft measurement model of the void fraction is established. By comparing with traditional recursive augmented least squares (RELS), particle swarm optimization (PSO), and simulated annealing-based PSO, the void fraction soft measurement model is identified and calculated separately. The segmentation optimization results of PSO based on simulated annealing have higher accuracy and stability than RELS and PSO, but as the number of center angles increases, the relative accuracy and stability of the system will deteriorate. And the characteristic is not conducive to the calculation and analysis of data results. By combining the actual model, the convolutional neural network weight update algorithm is added to the LSTM, and the RNN-LSTM convolutional neural network is used to predict the void fraction of the second half the region. It improves the effect of RNN gradient problem on learning ability and improves learning ability. Through comparison, it is found that the convolutional neural network based on RNN-LSTM has a better prediction effect, improves the accuracy and stability of the system, and provides a new method for the measured void fraction of twophase flow.

Complex artificial neural network with applications to wireless communications

We have developed a complex artificial neural network based radio receiver technology. This is based on novel complex neurons and deep networks of them. This type of radio does not have MIMO decoder or demodulator in a classical sense but instead the radio learns to demodulate through training - and can potentially adapt to jamming through a learning process. This type of radio can continuously adapt to changing environment and can potentially increase interference resistance spectral efficiency of modern radio networks.Complex neurons can natively operate on complex numbers, including multiplying (scale, rotation) of the input data. These networks, however, have proven chaotic and the convergence has proven to be highly problematic. The difficulty of determining complex coefficients, i.e. training, has made the practical use of these networks difficult or impossible.In this research, a new type of neuron is used with a network architecture that avoids most of the issues plaguing the convergence and training of the network. Combination of a new neuron weight update algorithm and the use of layers with nonlinear selection functions and methods to avoid large single error values in backpropagation allow complex neural network to be trained with fast convergence.

Delayed LMS Algorithm for Ballistocardiogram Biomedical Signal

The Ballisto-cardiogram (BCG) is a biomedical signal which is basically a measure of ballistic forces on heart. Just like ECG to detect the abnormalities in heart Ballisto-cardiography technique is used extensively now a day in research to analyze the abnormalities of the patient. When the blood pumps from heart to different parts of the body is represented in the form of graph for each heart beat. The frequency of Ballisto-cardiography signal is 1-20Hz .Ballisto-cardiography is most emerging techniques which is used to test the diseases related to heart called as cardiovascular disease. Various devices [1] like chairs, beds and weighing scales are projected to improvise the extraction of the BCG, but noise is one of the main issues with this BCG signal processing, noise is generated because of motion artifacts, shaking of the devices or may be power line noise. This noise [2] affects the quality of signal which we need to test. In order to overcome such issue this paper proposes a new architecture making use of LMS filtering algorithm Here weight update algorithm is used to update the error in extracted signal. The architecture proposed here includes FIR filter and also error computation blocks. Here the author has implemented 5-tap filtering algorithm. MATLAB and system generators are used to carry out the work.

Family of chaotic maps from game theory

ABSTRACT From a two-agent, two-strategy congestion game where both agents apply the multiplicative weights update algorithm, we obtain a two-parameter family of maps of the unit square to itself. Interesting dynamics arise on the invariant diagonal, on which a two-parameter family of bimodal interval maps exhibits periodic orbits and chaos. While the fixed point b corresponding to a Nash equilibrium of such map f is usually repelling, it is globally Cesàro attracting on the diagonal, that is, for every . This solves a known open question whether there exists a ‘natural’ nontrivial smooth map other than with centres of mass of all periodic orbits coinciding. We also study the dependence of the dynamics on the two parameters.

LrGAN: A Compact and Energy Efficient PIM-Based Architecture for GAN Training

As a powerful unsupervised learning method, Generative Adversarial Network (GAN) plays an essential role in many domains. However, training a GAN imposes four more challenges: (1) intensive communication caused by complex train phases of GAN; (2) much more ineffectual computations caused by peculiar convolutions; (3) more frequent off-chip memory accesses for exchanging intermediate data between the generator and the discriminator; and (4) high energy consumption of unnecessary fine-grained MLC programming. In this article, we propose LrGAN, a PIM-based GAN accelerator, to address the challenges of training GAN. We first propose a zero-free data reshaping scheme for ReRAM-based PIM, which removes the zero-related computations. We then propose a 3D-connected PIM, which can reconfigure connections inside PIM dynamically according to dataflows of propagation and updating. After that, we propose an approximate weight update algorithm to avoid unnecessary fine-grain MLC programming. Finally, we propose LrGAN based on these three techniques, providing different levels of accelerating GAN for programmers. Experiments show that LrGAN achieves 47.2×, 21.42×, and 7.46× speedup over FPGA-based GAN accelerator, GPU platform, and ReRAM-based neural network accelerator respectively. Besides, LrGAN achieves 13.65×, 10.75×, and 1.34× energy saving on average over GPU platform, PRIME, and FPGA-based GAN accelerator, respectively.

Adaptive Weight Update Algorithm for Target Tracking of UUV Based on Improved Gaussian Mixture Cubature Kalman Filter

The Gaussian mixture filter can solve the non-Gaussian problem of target tracking in complex environment by the multimode approximation method, but the weights of the Gaussian component of the conventional Gaussian mixture filter are only updated with the arrival of the measurement value in the measurement update stage. When the nonlinear degree of the system is high or the measurement value is missing, the weight of the Gauss component remains unchanged, and the probability density function of the system state cannot be accurately approximated. To solve this problem, this paper proposes an algorithm to update adaptive weights for the Gaussian components of a Gaussian mixture cubature Kalman filter (CKF) in the time update stage. The proposed method approximates the non-Gaussian noise by splitting the system state, process noise, and observation noise into several Gaussian components and updates the weight of the Gaussian components in the time update stage. The method contributes to obtaining a better approximation of the posterior probability density function, which is constrained by the substantial uncertainty associated with the measurements or ambiguity in the model. The estimation accuracy of the proposed algorithm was analyzed using a Taylor expansion. A series of extensive trials was performed to assess the estimation precision corresponding to various algorithms. The results based on the data pertaining to the lake trial of an unmanned underwater vehicle (UUV) demonstrated the superiority of the proposed algorithm in terms of its better accuracy and stability compared to those of conventional tracking algorithms, along with the associated reasonable computational time that could satisfy real-time tracking requirements.

Neural Computing Enhanced Parameter Estimation for Multi-Input and Multi-Output Total Non-Linear Dynamic Models

In this paper, a gradient descent algorithm is proposed for the parameter estimation of multi-input and multi-output (MIMO) total non-linear dynamic models. Firstly, the MIMO total non-linear model is mapped to a non-completely connected feedforward neural network, that is, the parameters of the total non-linear model are mapped to the connection weights of the neural network. Then, based on the minimization of network error, a weight-updating algorithm, that is, an estimation algorithm of model parameters, is proposed with the convergence conditions of a non-completely connected feedforward network. In further determining the variables of the model set, a method of model structure detection is proposed for selecting a group of important items from the whole variable candidate set. In order to verify the usefulness of the parameter identification process, we provide a virtual bench test example for the numerical analysis and user-friendly instructions for potential applications.