Search

In the construction process of radial basis function (RBF) networks, two common crucial issues arise: the selection of RBF centers and the effective utilization of the given source without encountering the overfitting problem. Another important issue is the fault tolerant capability. That is, when noise or faults exist in a trained network, it is crucial that the network's performance does not undergo significant deterioration or decrease. However, without employing a fault tolerant procedure, a trained RBF network may exhibit significantly poor performance. Unfortunately, most existing algorithms are unable to simultaneously address all of the aforementioned issues. This paper proposes fault tolerant training algorithms that can simultaneously select RBF nodes and train RBF output weights. Additionally, our algorithms can directly control the number of RBF nodes in an explicit manner, eliminating the need for a time-consuming procedure to tune the regularization parameter and achieve the target RBF network size. Based on simulation results, our algorithms demonstrate improved test set performance when more RBF nodes are used, effectively utilizing the given source without encountering the overfitting problem. This paper first defines a fault tolerant objective function, which includes a term to suppress the effects of weight faults and weight noise. This term also prevents the issue of overfitting, resulting in better test set performance when more RBF nodes are utilized. With the defined objective function, the training process is designed to solve a generalized M-sparse problem by incorporating an ℓ 0 -norm constraint. The ℓ 0 -norm constraint allows us to directly and explicitly control the number of RBF nodes. To address the generalized M-sparse problem, we introduce the noise-resistant iterative hard thresholding (NR-IHT) algorithm. The convergence properties of the NR-IHT algorithm are subsequently discussed theoretically. To further enhance performance, we incorporate the momentum concept into the NR-IHT algorithm, referring to the modified version as "NR-IHT-Mom". Simulation results show that both the NR-IHT algorithm and the NR-IHT-Mom algorithm outperform several state-of-the-art comparison algorithms., Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2024 Elsevier Ltd. All rights reserved.)

The concept of randomized neural networks (RNNs), such as the random vector functional link network (RVFL) and extreme learning machine (ELM), is a widely accepted and efficient network method for constructing single-hidden layer feedforward networks (SLFNs). Due to its exceptional approximation capabilities, RNN is being extensively used in various fields. While the RNN concept has shown great promise, its performance can be unpredictable in imperfect conditions, such as weight noises and outliers. Thus, there is a need to develop more reliable and robust RNN algorithms. To address this issue, this paper proposes a new objective function that addresses the combined effect of weight noise and training data outliers for RVFL networks. Based on the half-quadratic optimization method, we then propose a novel algorithm, named noise-aware RNN (NARNN), to optimize the proposed objective function. The convergence of the NARNN is also theoretically validated. We also discuss the way to use the NARNN for ensemble deep RVFL (edRVFL) networks. Finally, we present an extension of the NARNN to concurrently address weight noise, stuck-at-fault, and outliers. The experimental results demonstrate that the proposed algorithm outperforms a number of state-of-the-art robust RNN algorithms., Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2024 Elsevier Ltd. All rights reserved.)

Accurate 3-D mapping of multiple bioelectric sources in nerve fibers with high spatial resolution is challenging for the diagnosis and treatment of a variety of neural abnormalities. Ultrasound current source density imaging exploits the acoustoelectric (AE) effect, an interaction between electrical current and acoustic pressure waves propagating through a conducting material, and has distinct advantages over conventional electrophysiology (i.e., without ultrasound) for mapping electrical current flow in tissue. Ultrasound current source density imaging and two complementary Wheatstone bridge circuits were used to simultaneously detect two separate current flows induced in tissue phantoms. It has been found that the addition and subtraction of AE signals acquired by two circuits are independent components, regardless of whether the two sources are positioned at the same or different depths. In the ultrasound field, the AE signal from the bridge circuits is stronger, with a higher signal-to-noise ratio, than without a bridge circuit. Both experimental and simulated AE images depend on the magnitude and direction of the current, as well as the geometry (shape and thickness) and location of the current sources in the ultrasound field (2.25-MHz transducer). The experimental results are consistent with simulations consisting of multiple current sources. Real-time 3-D ultrasound current source density images of multiple current flows co-registered with convention pulse echo ultrasound potentially facilitate monitoring of neurologic disorders., (Copyright © 2016 World Federation for Ultrasound in Medicine & Biology. Published by Elsevier Inc. All rights reserved.)

Spatial discriminations can be performed using either egocentric information based on body position or allocentric information based on the position of landmarks in the environment. Beagle dogs ranging from 2 to 16 years of age were tested for their ability to learn a novel egocentric spatial discrimination task that used two identical blocks paired in three possible spatial positions (i.e. left, center and right). Dogs were rewarded for responding to an object furthest to either their left or right side. Therefore, when the center location was used, it was correct on half of the trials and incorrect on the other half. Upon successful acquisition of the task, the reward contingencies were reversed, and the dogs were rewarded for responding to the opposite side. A subset of dogs was also tested on an allocentric spatial discrimination task, landmark discrimination. Egocentric spatial reversal learning and allocentric discrimination learning both showed a significant age-dependent decline, while initial egocentric learning appeared to be age-insensitive. Intra-subject correlation analyses revealed a significant relationship between egocentric reversal learning and allocentric learning. However, the correlation only accounted for a small proportion of the variance, suggesting that although there might be some common mechanism underlying acquisition of the two tasks, additional unique neural substrates were involved depending on whether allocentric or egocentric spatial information processing was required.

In the classical deterministic Elman model, the estimation of parameters must be very accurate. Otherwise, the system performance is very poor. To improve the system performance, we can use a Kalman filtering algorithm to guide the operation of a trained recurrent neural network (RNN). In this case, during training, we need to estimate the state of hidden layer, as well as the weights of the RNN. This paper discusses how to use the dual extended Kalman filtering (DEKF) for this dual estimation and how to use our proposing DEKF for removing some unimportant weights from a trained RNN. In our approach, one Kalman algorithm is used for estimating the state of the hidden layer, and one recursive least square (RLS) algorithm is used for estimating the weights. After training, we use the error covariance matrix of the RLS algorithm to remove unimportant weights. Simulation showed that our approach is an effective joint-learning-pruning method for RNNs under the online operation.

The recursive least squared (RLS) algorithm is an effective online training method for neural networks. However, its conjunctions with weight decay and pruning have not been well studied. This paper elucidates how generalization ability can be improved by selecting an appropriate initial value of the error covariance matrix in the RLS algorithm. Moreover, how the pruning of neural networks can be benefited by using the final value of the error covariance matrix will also be investigated. Our study found that the RLS algorithm is implicitly a weight decay method, where the weight decay effect is controlled by the initial value of the error covariance matrix; and that the inverse of the error covariance matrix is approximately equal to the Hessian matrix of the network being trained. We propose that neural networks are first trained by the RLS algorithm and then some unimportant weights are removed based on the approximate Hessian matrix. Simulation results show that our approach is an effective training and pruning method for neural networks.

Although the performance of trellis coded vector quantizers (TCVQs) is much better than that of conventional memoryless vector quantizers, designing the trellis in TCVQs has a certain amount of computational overhead. In this paper, we apply Kohonen maps to simplify the design process of the TCVQs by exploiting the ordering property of Kohonen maps. This new implementation of the TCVQs is named as trellis coded Kohonen maps (TCKMs). Simulation results show that, with such a simple design process, the performance of the TCKMs is still comparable to that of the conventional TCVQs.