Characterising stimulus-specific adaptation using a multi-layer field model.

The response of an auditory neuron to a tone is often affected by the context in which the tone appears. For example, when measuring the response to a random sequence of tones, frequencies that appear rarely elicit a greater number of spikes than those that appear often. This phenomenon is called stimulus-specific adaptation (SSA). This article presents a neural field model in which SSA arises through selective adaptation to the frequently-occurring inputs. Formulating the network as a field model allows one to obtain an analytical expression for the expected response of a simple two-layer model to tones in a random sequence. The sequences of stimuli used in SSA experiments contain hundreds-and sometimes thousands-of tones, and these experiments routinely measure the response to many such sequences. A conventional neural network model (e.g., integrate-and-fire) would require numerical integration over long time periods to obtain results. Consequently, a field model that offers an immediate, analytical solution for a given input sequence is helpful. Two routes to obtaining this solution are discussed. The first involves the convolution of two closed-form expressions; the second relies on a series of approximations involving Gaussian curves. The purpose of the paper is to describe the model, to develop the approximations that allow an analytical solution, and finally, to comment on the output of the model in light of the SSA results published in the physiology literature. This article is part of a Special Issue entitled "Neural Coding".