Advances in neural network based learning

Abstract

In recent years, the applications of diverse machine leaning algorithms and their fusion to the cybernetics and decisionmaking have been attracting more and more scholars from different disciplines. As a most frequently used technique of extracting knowledge from data, a feed-forwarded neural network (NN) has shown more advantages for the regression and classification problems. The main advantages come from the capability of good approximation of NNs and their corresponding highly nonlinear boundary. Due to the many merits and wide applications of NNs, the study on several fundamental issues of NNs, such as structure selection, overtraining, robustness and capability of resistance to noisy data, training on different types of data, and more importantly, the generalization ability, are still in progress although NNs have been an old topic in the areas of learning and reasoning. As a result, these extensive studies and significant improvements bring a number of new features anddevelopments to NNs. This issue makes an attempt to provide some latest advances of NN learning, the recent improvements of performance for NN learning systems, and new applications of NNs to different real fields. In this issue, 14 papers are accepted for publication. The 14 papers cover a variety of topics which include the stochastic stability NN with time delay, imbalanced and uncertain data, fuzzy NNs, bidirectional associative memory (BAM), deep learning of NNs, and other comparative and review studies. Categorization and brief description of these papers are given below. Four papers discuss the stability of NNs with delay time. Actually, the time delay is ubiquitous in most natural systems. Since the time delay is frequently encountered in NNs, the issue about stability of NNs with delay has been emerging as a challenging task for NN researchers. The paper authored by Cheng-De Zheng studied a class of stochastic reaction diffusion NNs with Markovian jumping parameters and time delays. Lyapunov functional method and stochastic analysis technique are used there to construct a new stochastic stability in the principle of the mean square. Continuing this topic, the paper authored by M. Syed Ali studied the stochastic stability analysis of uncertain recurrent NNs, also with Markovian jumping parameters and time varying delays. By using Lyapunov functional theory, Ito differential rule and matrix analysis techniques, a sufficient criterion in term of linear matrix inequality (LMI) is established such that, for all admissible parameter uncertainties and stochastic disturbances, the stochastic NN is perfectly stable. And then, a non-stationary problem on BAM NNs is discussed in Qingqing He’s paper. An exponential stability in terms of LMIs is achieved for a class of interval C-G type (BAM) NNs with the mixed time delays and non-smooth behaved functions by using Homomorphic mapping theory, non-smooth analysis, LMIs, free-weighting matrix and Lyapunov–krasovskii functional approach. Furthermore, the paper authored by R. Raja derived the stability analysis for a class of uncertain discrete-time stochastic BAM NNs with timevarying delays and impulses. Lyapunov–Krasovskii functional and stochastic analysis techniques were applied to obtain a set of sufficient conditions in term of LMI. Along this line, the non-stationary challenge in other class of NN can be tackled. The paper authored by Adel X.-Z. Wang A. B. Musa (&) Key Laboratory in Machine Learning and Computational Intelligence of Hebei Province, College of Mathematics and Computer Science, Hebei University, Baoding 071002, China e-mail: abdubashir20@yahoo.com