Boudour Ammar

Existence and Uniqueness of Pseudo Almost-Periodic Solutions of Recurrent Neural Networks With Time-Varying Coefficients and Mixed Delays

This paper is concerned with the existence and uniqueness of pseudo almost-periodic solutions to recurrent delayed neural networks. Several conditions guaranteeing the existence and uniqueness of such solutions are obtained in a suitable convex domain. Furthermore, several methods are applied to establish sufficient criteria for the globally exponential stability of this system. The approaches are based on constructing suitable Lyapunov functionals and the well-known Banach contraction mapping principle. Moreover, the attractivity and exponential stability of the pseudo almost-periodic solution are also considered for the system. A numerical example is given to illustrate the effectiveness of our results.

Biped robot control using particle swarm optimization

In this paper we propose a method to generate gaits of a biped robot by a particle swarm optimization algorithm. The system generates angular positions for joints with an interpolate end segments positions to evaluate walking stability. The proposed PSO is adapted to generate angular position joints, Human walking stability criteria are used to check and validate the gaits. The experimental procedure includes a robot assembly and online test. Then an upper torso controller is introduced to correct walking stability and limits fall downs.

PSO-based analysis of Echo State Network parameters for time series forecasting

Graphical abstractDisplay Omitted HighlightsEcho State Network (ESN) is an interesting tool for dealing with time series forecasting problems.The learning performance of ESN can be affected because of the random setting of some untrained weights.PSO is introduced to ESN as a pre-training tool to optimize the untrained weights.The networks weights become suitable to the targeted application.The accuracy of ESN is considerably improved after PSO pre-training. Echo State Networks, ESNs, are standardly composed of additive units undergoing sigmoid function activation. They consist of a randomly recurrent neuronal infra-structure called reservoir. Coming up with a good reservoir depends mainly on picking up the right parameters for the network initialization. Human expertise as well as repeatedly tests may sometimes provide acceptable parameters. Nevertheless, they are non-guaranteed. On the other hand, optimization techniques based on evolutionary learning have proven their strong effectiveness in unscrambling optimal solutions in complex spaces. Particle swarm optimization (PSO) is one of the most popular continuous evolutionary algorithms. Throughout this paper, a PSO algorithm is associated to ESN to pre-train some fixed weights values within the network. Once the networks initial parameters are set, some untrained weights are selected for optimization. The new weights, already optimized, are re-squirted to the network which launches its normal training process. The performances of the network are a subject of the error and the time processing evaluation metrics. The testing results after PSO pre-training are compared to those of ESN without optimization and other existent approaches. The conceived approach is tested for time series prediction purpose on a set of benchmarks and real-life datasets. Experimental results show obvious enhancement of ESN learning results.

Learning to Walk Using a Recurrent Neural Network with Time Delay

Walking based on gaits is one of the most approved methodologies for walking robot. In this paper, we develop learning strategy for walking biped robot or human based on a self made database using biomechanical capture. This system is provided by a Recurrent Neural Network (RNN) with an internal discrete time delay. The role of the proposed network is the training of human walking data by giving an estimation of the bipeds next position at each time and achieve a human-like natural walking. Different architectures of RNN are proposed and tested. In Particular, a comparative study is given and the results of the RNN mixed with extended Kalman filter are illustrated.

On the dynamics of the high-order type of neural networks with time varying coefficients and mixed delay

This paper discuss the oscillations of high-Order type recurrent delayed neural networks. Various creteria are used to prove the existence and uniqueness of pseudo almost periodic solution in a suitable convex domain. Our method is based on constructing suitable Lyapunov functionals and the well-known Banach contraction mapping principle. Banach fixed point, pseudo almost-periodic functions, high order recurrent neural network.

Learning system for standing human detection

The human detection is a key functionality to reach Human Computer and Robot Interaction. The human tracking is also a rapidly evolving area in computer and robot vision; it aims to explore and to follow human motion. We present in this article an intelligent system to learn human detection. The descriptors used in our system make up the combination of HOG and SIFT that capture salient features of humans automatically. Additionally, this system is employed to follow humans that it detects. Experimental results have been extracted for a set of sequences with standing and moving people at different positions and with a variation of backgrounds.

Single- and multi-objective particle swarm optimization of reservoir structure in Echo State Network.

Echo State Networks ESNs are specific kind of recurrent networks providing a black box modeling of dynamic non-linear problems. Their architecture is distinguished by a randomly recurrent hidden infra-structure called dynamic reservoir. Coming up with an efficient reservoir structure depends mainly on selecting the right parameters including the number of neurons and connectivity rate within it. Despite expertise and repeatedly tests, the optimal reservoir topology is hard to be determined in advance. Topology evolving can provide a potential way to define a suitable reservoir according to the problem to be modeled. This last can be mono- or multi-constrained. Throughout this paper, a mono-objective as well as a multi-objective particle swarm optimizations are applied to ESN to provide a set of optimal reservoir architectures. Both accuracy and complexity of the network are considered as objectives to be optimized during the evolution process. These approaches are tested on various benchmarks such as NARMA and Lorenz time series.

Incremental learning approach for human detection and tracking

Human detection is a key functionality to reach Human Robot/Computer Interaction. The human tracking is also a rapidly evolving area in computer and robot vision; it aims to explore and to follow human motion. We present in this article an intelligent system to learn human detection. The descriptors used in our system make up the combination of HOG and SIFT that capture salient features of humans automatically. Additionally, an incremental PCA is employed to follow the detected humans. Experimental results have been extracted for a set of sequences with standing and moving people at different positions and with a variation of backgrounds.

Stability and Exponential Synchronization of High-Order Hopfield Neural Networks with Mixed Delays

ABSTRACT This paper investigates the problems of stability and synchronization for high-order recurrent neural networks with mixed delays. Firstly, we establish sufficient conditions to ensure the asymptotic stability and then the exponential synchronization. Furthermore, our results are applied to two chosen systems to demonstrate the effectiveness of the obtained theoretical results.

A Hybrid Approach Based on Particle Swarm Optimization for Echo State Network Initialization

Echo state networks (ESNs) fulfill considerable promises for topology fine-tuning in supervised training. However the randomness of the setting of ESN weights initialization affects badly the learning performance. On the other side, Particle Swarm Optimization (PSO) has proven its efficiency as an optimization tool to puzzle out optimal solutions in complex space. In this work, we present an ESN architecture to which we associate a PSO algorithm to pre-train the weights within the network layers. A random distribution of the weights matrices is firstly performed. Then, these weights are pre-trained in order to fit the application requirements. Once optimized, they are re-injected into the ESN model which, in its turn, undergoes a training process followed by a test phase. A comparison between the network performances before and after optimization process is performed. Empirical results show a reduction of learning errors in the case of PSO use.