Temporal Information Transformed into a Spatial Code by a Neural Network with Realistic Properties

Abstract

Neurons exhibit a wide range of properties in addition to postsynaptic potential (PSP) summation and spike generation. Although other neuronal properties such as paired-pulse facilitation (PPF) and slow PSPs are well characterized, their role in information processing remains unclear. It is possible that these properties contribute to temporal processing in the range of hundreds of milliseconds, a range relevant to most complex sensory processing. A continuous-time neural network model based on integrate-and-fire elements that incorporate PPF and slow inhibitory postsynaptic potentials (IPSPs) was developed here. The time constants of the PPF and IPSPs were estimated from empirical data and were identical and constant for all elements in the circuit. When these elements were incorporated into a circuit inspired by neocortical connectivity, the network was able to discriminate different temporal patterns. Generalization emerged spontaneously. These results demonstrate that known time-dependent neuronal properties enable a network to transform temporal information into a spatial code in a self-organizing manner--that is, with no need to assume a spectrum of time delays or to custom-design the circuit.