Li and Atick's theory of efficient binocular coding: A tutorial and mini-review

Abstract

Li and Atick (Network: Computation in Neural Systems 5 (1994) 157-174) presented a theory of efficient binocular encoding that explains a number of experimental findings. A binocular neuron is conventionally described in terms of two channels: the left and right eyes. Li and Atick's theory instead describes the neuron in terms of two alternative channels: the binocular sum and difference. The advantage of the latter description is that, unlike the left and right eye channels, the summation and differencing channels are usually uncorrelated; this means that each channel can be optimised independently of the other. The theory shows how to derive optimal receptive fields for the binocular summation and differencing channels; from these, it is easy to derive the neuron's optimal left and right eye receptive fields. The functional reality of the summation and differencing channels is demonstrated by a series of adaptation studies that confirm some counterintuitive predictions of the theory. Here we provide an accessible account of the theory, and review the evidence supporting it.

Keywords: Binocular vision; Efficient coding; Theoretical neuroscience.