Recognition of temporal event sequences by a network of cortical neurons

Recognition of ordered sequences of temporal events is central to many perceptual recognition tasks, from speech detection to analysis of biological motion. We describe a simple cortical network capable of recognizing event sequences through a process of encoding followed by detection. The network is composed of regular spiking and fast spiking neurons, with minimal connectivity. Ordered sequences of inputs occurring over tens-to-hundreds of milliseconds, are time compressed by the network into tightly clustered spike outputs occurring over a few milliseconds. We investigate the ability of the network to accurately encode the input pattern, in the presence or absence of noise. We show that information about relative input timings are preserved in the output interspike intervals.