MELM-GRBF: A modified version of the extreme learning machine for generalized radial basis function neural networks

In this paper, we propose a methodology for training a new model of artificial neural network called the generalized radial basis function (GRBF) neural network. This model is based on generalized Gaussian distribution, which parametrizes the Gaussian distribution by adding a new parameter τ. The generalized radial basis function allows different radial basis functions to be represented by updating the new parameter τ. For example, when GRBF takes a value of τ=2, it represents the standard Gaussian radial basis function. The model parameters are optimized through a modified version of the extreme learning machine (ELM) algorithm. In the methodology proposed (MELM-GRBF), the centers of each GRBF were taken randomly from the patterns of the training set and the radius and τ values were determined analytically, taking into account that the model must fulfil two constraints: locality and coverage. An thorough experimental study is presented to test its overall performance. Fifteen datasets were considered, including binary and multi-class problems, all of them taken from the UCI repository. The MELM-GRBF was compared to ELM with sigmoidal, hard-limit, triangular basis and radial basis functions in the hidden layer and to the ELM-RBF methodology proposed by Huang et al. (2004) [1]. The MELM-GRBF obtained better results in accuracy than the corresponding sigmoidal, hard-limit, triangular basis and radial basis functions for almost all datasets, producing the highest mean accuracy rank when compared with these other basis functions for all datasets.     

Global exponential stability in Lagrange sense for neutral type recurrent neural networks

In this paper, the global exponential stability in Lagrange sense for continuous neutral type recurrent neural networks (NRNNs) with multiple time delays is studied. Three different types of activation functions are considered, including general bounded and two types of sigmoid activation functions. By constructing appropriate Lyapunov functions, some easily verifiable criteria for the ultimate boundedness and global exponential attractivity of NRNNs are obtained. These results can be applied to monostable and multistable neural networks as well as chaos control and chaos synchronization.     

Phase synchronizing in Hindmarsh-Rose neural networks with delayed chemical coupling

Although diffusive electrical connections in neuronal networks are instantaneous, excitatory/inhibitory couplings via chemical synapses encompass a transmission time-delay. In this paper neural networks with instantaneous electrical couplings and time-delayed excitatory/inhibitory chemical connections are considered and scaling of the spike phase synchronization with the unified time-delay in the network is investigated. The findings revealed that in both excitatory and inhibitory chemical connections, the phase synchronization could be enhanced by introducing time-delay. The role of the variability of the neuronal external current in the phase synchronization is also investigated. As individual neuron models, Hindmarsh-Rose model is adopted and the network structure of the electrical and chemical connections is considered to be Watts-Strogatz and directed random networks, respectively.     

Stability and synchronization for Markovian jump neural networks with partly unknown transition probabilities

This paper addresses the problems of stability and synchronization for a class of Markovian jump neural networks with partly unknown transition probabilities. We first study the stability analysis problem for a single neural network and present a sufficient condition guaranteeing the mean square asymptotic stability. Then based on the Lyapunov functional method and the Kronecker product technique, the chaos synchronization problem of an array of coupled networks is considered. Both the stability and the synchronization conditions are delay-dependent, which are expressed in terms of linear matrix inequalities. The effectiveness of the developed methods is shown by simulation examples.     

Mean square exponential stability in high-order stochastic impulsive BAM neural networks with time-varying delays

In this paper, a class of stochastic impulsive high-order BAM neural networks with time-varying delays is considered. By using Lyapunov functional method, LMI method and mathematics induction, some sufficient conditions are derived for the globally exponential stability of the equilibrium point of the neural networks in mean square. It is believed that these results are significant and useful for the design and applications of impulsive stochastic high-order BAM neural networks.     

Global exponential stability of neural networks with discrete and distributed delays and general activation functions on time scales

By employing time scale calculus theory, free weighting matrix method and linear matrix inequality (LMI) approach, several delay-dependent sufficient conditions are obtained to ensure the existence, uniqueness and global exponential stability of the equilibrium point for the neural networks with both infinite distributed delays and general activation functions on time scales. Both continuous-time and discrete-time neural networks are described under the same framework by the reported method. Illustrated numerical examples are given to show the effectiveness of the theoretical analysis. It is noteworthy that the activation functions are assumed to be neither bounded nor monotone.     

Stability analysis for discrete delayed Markovian jumping neural networks with partly unknown transition probabilities

This paper addresses the analysis problem of asymptotic stability for a class of uncertain neural networks with Markovian jumping parameters and time delays. The considered transition probabilities are assumed to be partially unknown. The parameter uncertainties are considered to be norm-bounded. A sufficient condition for the stability of the addressed neural networks is derived, which is expressed in terms of a set of linear matrix inequalities. A numerical example is given to verify the effectiveness of the developed results.     

Outliers detection in environmental monitoring databases

Environmental monitoring is nowadays an important task in many industrial operations. In order to comply with strong environmental laws, they have implemented monitoring systems based on a network of air quality and meteorological stations providing real-time measurements of key variables associated to the distribution of pollutants in surrounding areas. These measurements can be contaminated by outliers, which must be discarded in order to have a consistent set of data. This work presents a nonlinear procedure for outliers detection based on residual analysis of regression with Partial Least Squares and Artificial Neural Networks. In order to minimize the negative effect of outliers in the training dataset a learning algorithm with regularization is proposed. This algorithm is based on a Quasi-Newton optimization method and it was tested on a simulated nonlinear process, on real data from environmental monitoring contaminated with synthetic outliers, and finally applied to a real environmental monitoring data obtained from a monitoring station and having natural outliers. The results are encouraging and further developments are foreseen for including information from neighboring stations and emission source operation.     

An agent-based signal processing in-node environment for real-time human activity monitoring based on wireless body sensor networks

Nowadays wireless body sensor networks (WBSNs) have great potential to enable a broad variety of assisted living applications such as human biophysical/biochemical control and activity monitoring for health care, e-fitness, emergency detection, emotional recognition for social networking, security, and highly interactive games. It is therefore important to define design methodologies and programming frameworks which enable rapid prototyping of WBSN applications. Several effective application development frameworks have been already proposed for WBSNs designed for TinyOS-based sensor platforms, e.g. CodeBlue, SPINE, and Titan. In this paper we present an application of MAPS, an agent framework for wireless sensor networks based on the Java-programmable Sun SPOT sensor platform, for the development of a real-time WBSN-based system for human activity monitoring. The agent-oriented programming abstractions provided by MAPS allow effective and rapid prototyping of the sensor-side software. In particular, the architecture of the developed system is a typical star-based WBSN composed of a coordinator node and two sensor nodes located respectively on the waist and the thigh of the monitored assisted living. The coordinator relies on a JADE-based enhancement of the SPINE coordinator and allows configuring sensors, receiving their data, and recognizing pre-defined human activities. On the other hand, each sensor node runs a MAPS-based agent that performs sensing of the 3-axial accelerometer sensor, computation of significant features on the acquired data, feature aggregation and transmission to the coordinator. The experimentation phase of the prototype, which allows evaluating the obtainable monitoring performances and activity recognition accuracy, is described. Moreover, a comparison of the monitoring system based on MAPS, AFME and SPINE in terms of programming effectiveness and system performances is discussed.