Design of continuous-time recurrent neural networks with piecewise-linear activation function for generation of prescribed sequences of bipolar vectors (Q6057957)

Recurrent neural networks (RNNs) have been studied for many years due to their many important applications in subjects varying from combinatorial optimization, image processing, acoustic modeling, language processing and many more. We refer the reader to the paper for an extensive literature review in this regard and also for a excellent reference list regarding recent advances in the study and applications of RNNs. This interesting paper focuses on a continuous-time RNN model with a piecewise-linear activation function that has neither external inputs nor hidden neurons, and studies the problem of finding the parameters of the model so that it generates a given sequence of bipolar vectors. The authors provide a sufficient condition for their model to generate their desired sequence of bipolar vectors which is expressed as a system of linear inequalities in the parameters. The authors provide three interesting approaches to finding solutions of this system of linear inequalities. One approach is via a convex quadratic programming method and the other two via linear programming problems. As a result, two types of sequences of bipolar vectors for their model are generated. The case where the model generates a periodic sequence of bipolar vectors is also studied, and a sufficient condition for the trajectory of the state vector to converge to a limit cycle is given. This paper is well written, with ample references and presents a nice theoretical framework to solve a difficult and interesting problem.