Time computation of spiking networks for world-centered representation

This article argues that full-parallel algorithms of traditional sigmoidal neural networks based on average fire rate which has excluded useful time operation seem improper for the invariance. A speaking network that introduces a space search mechanism into neural computing is presented. The search can transform space coordinate into time coordinate. Consequently, the operations based relative positions are converted into those based relative time which are very easy for neural realization through delay connections. Accordingly, a world-centered model (WCM) that consists of a space searcher, and a feature transfer vector (FTV) memory are developed. WCM is a pure neural network that represents the neural principle of invariant recognition.