Algorithms Design and Analysis for Binary Neural Networks and Cellular Nonlinear Networks

Description

This proposal aims to carry out efficient algorithms design for Binary Neural Networks (BNN) and Cellular Nonlinear Networks (CNN), with basic mathematical analysis on Cellular Automata in general, from the dynamical systems theory approach, targeting potential real-world applications in brain-like computing and new information processing technologies.First, the classical extended one-dimensional Cellular Automata (CA) will be combined with bi-infinite symbolic sequences to perform rigorous mathematical analysis, and the rich and complex dynamics of a class of Elementary Cellular Automata (ECA) rules will be analyzed from the symbolic dynamical systems viewpoint. Then, the behaviors of the perceptron of BNN will be characterized. Moreover, based on biological concepts and methods, several new algorithms will be designed for efficiently performing a huge number of Boolean functions. Finally, the designed algorithms will be developed into apiece of PC-software that can be easily programmed into hardware realization of BNN and CNN perceptrons.