NETWORK CONNECTIONS THAT EVOLVE TO CONTEND WITH THE INVERSE OPTICS PROBLEM

Abstract

A fundamental problem in vision is understanding how useful perceptions and behaviors arise in the absence of direct information about the physical sources of retinal stimuli (the inverse optics problem). With respect to light intensity, psychophysical studies of lightness and brightness show that humans contend with this problem by generating percepts that accord with the cumulative probabilities of naturally occurring luminance patterns (Yang and Purves, 2004). To understand the neural mechanisms underlying this strategy, we examined the connections of simple neural networks with four neurons (a target sensor, a context sensor, an integrating neuron and a response neuron) and three synaptic connections between them. The networks were presented with patterns of adjacent luminance values drawn from natural scenes and evolved to respond to the luminance of the target sensor according to the cumulative probabilities of the stimuli experienced. The networks had no information about the physical sources underlying the stimuli; nor did they perform feature analysis or represent images. The evolved responses were similar to human psychophysical functions of lightness magnitude estimation and contrast, and the evolved connectivity to lateral inhibition observed in biological circuitry. The evolved excitatory connection from the target sensor allows the network to non-linearly scale its responses, whereas the inhibitory connection from the context sensor results in contrast effects. These observations imply that animal vision uses the same general strategy and mechanisms to contend with the inherent inability of light stimuli to specify physical parameters in the environment.

Reference: Yang Z & Purves D (2004) The statistical structure of natural light patterns determines perceived light intensity. Proc Natl Acad Sci USA 101: 8745-8750.

Meeting abstract presented at VSS 2013