Top Research Papers on Neural Networks

A highly interactive, self-paced package, Neuralis, that can be used in Artificial Intelligence courses in general, and Neural Networks in particular, and its ease of operation and graphical interface, that allows for instant visualization of the whole process, are highlighted.

This work investigates several aspects of comparato r networks and shows that the min-max model is the ‘strongest’ model of computation which obeys the 0-1 principle.

This work formulation the hypernetwork training objective as a compromise between accuracy and diversity, where the diversity takes into account trivial symmetry transformations of the target network, and explains how it is related to variational inference.

The design of the Hopfield associative memory is reformulated in terms of a constraint satisfaction problem, and an electronic neural net capable of solving this problem in real time is proposed.

This paper introduces a real-time approach for generating a symbolic representation of the function learned by a neural network via another neural network (called the Interpretation Network, or I-Net), which maps network parameters to a symbolic representations of the network function.

Two normalised measures of neutrality based on a progressive random walk algorithm are proposed and shown to agree with visual analysis of two-dimensional benchmark problems, and are shown to scale well to higher dimensions.

The simulation results showed that the Neuro-PID controller controlled the pH at a desired setpoint with a better performance than the conventional PID controller.

The principle aim of this paper is to explore the existence of periodic solution of neural networks model with neutral delay by means of an abstract continuous theorem of k-set contractive operator and some analysis technique.

The stability of neutral cellular neural networks was studied based on linear matrix inequality technique and the result shows the validity of the presented algorithm is less conservative than that of Y.M.ZHANG et al.

It is found that Hopf bifurcation occurs when τ is across some critical values, and the global existence of periodic solution is established by using a global hopf bIfurcation result.

This paper introduces a previously unknown steganographic technique that can be exploited by adversaries if left unchecked and hides the existence of a secret neural network with arbitrary desired functionality within a carrier network.

A neural network includes a weight correction calculator that receives a desired output signal, determines a deviation of the neuron sum from the desiredoutput signal value, and modifies respective corrective weights using the determined deviation.

This work proposes a novel framework for designing an effective branching strategy for BaB, and learns a graph neural network (GNN) to imitate the strong branching heuristic behaviour.

The Hybrid model (ARIMA-ANN) was the best model in forecasting by stock index EGX30 and it is better than ARIMA and ANN which did singularly, that is because Hybrid Model has the minimum accurately values of forecasting standards.

The conclusions are of a generic nature, pointing out some pitfalls of neural network pruning in general, and of a more speciic nature, identifying the best pruning methods for high order perceptrons.

The simulation results showed that the neural network LSTM successfully coped with the task of predicting changes in insulation resistance, taking into account changing environmental factors, and will avoid sudden failures of electrical components, electric shock and fire caused by a decrease in insulation Resistance below the critical value.

The proposed idea of continuous monitoring of the insulation resistance with the prediction of its changes for a certain period will avoid sudden failures of electrical components, electric shock and fire caused by a decrease in insulation resistance below the critical value.

N2Net, a system that implements binary neural networks using commodity switching chips deployed in network switches and routers, shows that these devices can run simple neural network models at packet processing speeds of billions of packets per second.

An event-triggering scheme is proposed and utilized when the controller transfer signals to the complex networks, and some synchronization criteria are derived to ensure the global mean-square exponential synchronization of state trajectories of the remote master and the coupled dynamic networks.

Since the publication of Rumelhart and McClelland's seminal work on neural network theory, in 1986, the field has grown and grown as people come to realise the potential that neural networks have for dealing with complex pattern recognition problems.

Topographical studies on neuronal pathways are more and more completed by functional informations by using highly specific neurochemical and immunohistochemical techniques to determine the chemical nature of pathways.

This paper presents basics and brief about neural n etwork, artificial neural network (ANN), biological neuralnetwork (BNN) in soft computing, and NNs are useful for mapping problems.

To provide a neural semi-conductor chip including a neuron unit which is composed of plural neuron circuits to which a signal is supplied by various kinds of buses, an additional OR function is executed between the whole corresponding first global result signals and the global output signals by a dot processing in an off-chip common communication bus inside a driver block.

An improved structured DNN controller design that comprises a primary and a secondary neural network that provides stability guarantees, and the secondary neural network can be optimized to allow for improved adaptability to changing initial states.

An unsupervised method to learn a neural network, namely an explainer, to interpret a pre-trained convolutional neural network (CNN), i.e., explaining knowledge representations hidden in middle conv-layers of the CNN, which has significantly boosted the interpretability of CNN features.

The results indicate that ANNs are significantly better than Wiener models in predicting neural responses and are also able to predict responses of the same neuron in different individuals irrespective of which animal was used to develop the model.

This book is the first of a series of technical reports of a key research project of the Real-World Computing Program supported by the MITI of Japan to model human intelligence by a special class of mathematical systems called neural logic networks.

A novel framework leveraging Graph Neural Networks to Generate Neural Networks (GNN2GNN) where powerful NN architectures can be learned out of a set of available architecture-performance pairs, and paves the way towards generalisation between datasets.

This work proposes a novel machine learning framework that can be used for designing an effective branching strategy as well as for computing better lower bounds, and learns two graph neural networks that both directly treat the network they want to verify as a graph input and perform forward-backward passes through the GNN layers.

This paper introduces graphon NNs as limit objects of GNNs and proves a bound on the difference between the output of a GNN and its limit graphon-NN if the graph convolutional filters are bandlimited in the graph spectral domain.

This paper introduces Porcupine Neural Networks (PNNs) whose weight vectors are constrained to lie over a finite set of lines, and suggests that an unconstrained neural network can be approximated using a polynomially-large PNN.

This paper presents a new architecture of neural networks - intelligence increasing neural network (IINN), which achieves the building, memory and use of the data base with clear structure and indicates that when weak learners are independent, IINN has better performance than AdaBoost.

It is argued to distinguish rule extraction using neural networks and rule extraction for neural networks according to their different goals, where fidelity and accuracy should be excluded from the rule quality evaluation framework, respectively.

This work proposes a novel Calibration-Emulation-Sampling (CES) strategy to significantly enhance the computational efficiency of BNN and demonstrates that the proposed method improves computational efficiency of BNN, while maintaining similar performance in terms of prediction accuracy and uncertainty quantification.

The neural network inversion method is extended to the quaternion domain, which has the advantage of expressing 3D (three-dimensional) object attitudes easily and can define inverse problems where the cause and the result are expressed by the quaternity.

The Cortex Neural Network is an upper architecture of neural networks which motivated from cerebral cortex in the brain to handle different tasks in the same learning system and is able to identify different tasks and solve them with different methods.

The algorithm uses the clustering characteristic of ART neural network to design the RBF neural network structure, and the number of the hidden layer nodes and initial parameters are determined, so that the network has simplified structure.

A new kind of FNN: vague neuron network (VNN) is put forward, the properties are discussed and the VNN is applied to the problem of fault diagnosis and shows good performance.

The basic building blocks of the neural networks, their construction mechanisms and methodology for learning are discussed.

This analysis makes clear the strong similarity between linear neural networks and the general linear model developed by statisticians.

A new kind of FNN: vague neuron network (VNN) is put forward, the properties are discussed and the VNN is applied to the problem of fault diagnosis and shows good performance.

The application of neural networks to a different diagnostic problem, the diagnosis of faults in newly manufactured engines and the utility of Neural networks for process control are explored.

The designed model‐free online adaptive controller was implemented to a laboratory scaled pH process in real time by use of a dSPACE 1104 interfacing card and shows good tracking for both the set point and load changes over the entire nonlinear region.

By using Lyapunov-Krasovskii functional and Ito formula combining the linear matrix inequality method,exponential stability for stochastic neutral-type neural networks is discussed and exponentially stable criterion is given in mean square in this paper.

The delay-dependent stability criterion with full-order state estimator is proposed in terms of linear matrix inequalities and numerical example is given to illustrate the effectiveness of the proposed methods.

A new stability condition is presented based on the Lyapunov-Krasovskii method and the inequality technique, which is dependent on the amount of delay, in the form of a linear matrix inequality (LMI).

The proposed emotion detection system provides an improvement of approximately 10% using excitation source features and 3% using vocal tract system features over the recently proposed emotion detection which uses the energy and pitch contour modeling with functional data analysis.

An explanation of how neural networks can be applied to such important tasks as fault diagnosis and accommodation is presented and neural networks are shown to be part of the hierarchy of intelligent control where a higher order decision element monitors and supervises lower order elements for sensing and actuation.

Stability analysis of neural networks and systems with time delay in leakage (or “forgetting”) term and the delay decomposition approach was successfully introduced in [130] for the neural networks with constant delay.

Simulation results have shown the robustness and adaptability of the neural network PID control system of pH neutralization process and conventional PID control cannot adapt the changes of process.