Two Layered Synchronized Spiking Neural System for Feature Binding and Figure Recognition

摘要

One of the most attractive questions in the study of theoretical neurobiology is how visual information is carried and synthesized by the neuron spikes. Our model is based on the idea that synchronous firing encodes visual information. The system employs spike response neuron model. To design this neural system plausibly, we also concerned several neurophysiological issue related to the structure of mammalian visual cortex, i.e. the form of feed-forward and feed-back connections of inter-cortical neurons. The system consists of two layers: Lower layer is edge layer which simulates column neurons in the primary visual cortex. Edge layer composed of 7x7 retinal field. A retinal field includes 8 spiking neurons corresponding 8 directions. Higher layer is Feature linking layer. Feature linking layer is composed of oscillator neurons, each of which respectively corresponds to a peculiar figure. The weights of the neurons in the linking layer were previously trained by Hebbian learning rule. In the experiment, the system learned a square and a triangle figure. Correlative firing in the feature group is amplified via feed-back connection and coincident detector neuron. Experiment result shows that several bound feature groups can respectively encode the different objects simultaneously in 500msec order time scale.