Fast and adaptive network of spiking neurons for multi-view visual pattern recognition

In this paper, we describe and evaluate a new spiking neural network (SNN) architecture and its corresponding learning procedure to perform fast and adaptive multi-view visual pattern recognition. The network is composed of a simplified type of integrate-and-fire neurons arranged hierarchically in four layers of two-dimensional neuronal maps. Using a Hebbian-based training, the network adaptively changes its structure in order to respond optimally to different visual patterns. Neurons in the last layer accumulate information collected over multiple frames to reach a final decision. We tested the network with VidTimit dataset to recognize individuals using facial information from multiple frames. The experiments illustrate and evaluate the two main novelties of the network: structural adaptation and frame-by-frame accumulation of opinions.