Fast Neural Synchronization Using Geometrical Analysis for Random Walk in Search Space

Abstract

Neural synchronization is a phenomenon of mutual learning between two or more neural networks. This phenomenon has tremendous applications especially in cryptography where it can be efficiently used in key exchange over public and un secure networks. This biggest problem facing neural synchronization is the number of rounds needed to accomplish the synchronization; where the output of each neural is sent to other party over network. Since the synchronization is a stochastic behavior then it could be established anytime along the synchronization session, but no current approach to verify establishing the synchronization once it happened. This paper has deployed geometric analysis to investigate random walk of weights in plane during the synchronization session, the outcome is an enhancement of the convergence of neural networks to synchronization points and another outcome is the verification of establishment the synchronization which is the focus of this paper.