Spiking-timing based pattern recognition with real-world visual stimuli

Pattern recognition has been widely studied in the field of computational intelligence. However, primates outperform existing algorithms in cognitive tasks without any difficulty and most of current methods lack enough biological plausibility. Inspired by recent biological findings, a spike-timing based computational model is described, in which information is represented by temporal codes with explicit firing times rather than firing rates of neurons. Visual stimulation is converted into precisely timed spikes by a retina-like model. Encoded spatiotemporal patterns are learned by a temporal learning algorithm based on spiking-timing-dependent plasticity (STDP). The computational model integrates encoding and learning with a unified neural representation closing the gap between them. We show that our integrated model is capable of recognizing real world stimuli such as images successfully with fast and efficient neural computation.