Parallel double-conversion spiking neural network for world-centered recognition

The separation of patterns from ground is the necessary requirement for recognition. Most of neural recognition models are network-centered without the ability to extract patterns. As a result, some non-neural methods and the learning of pattern's variant positions are used to complete the task. This article presents a spiking double-conversion network (DCN) to search for patterns in input using the double conversions from the network-centered input vector to a time sequence and further from the sequence to pattern-centered vector. DCN is designed for network-centered recognition and cluster models to extend them to world-centered ones.