Sense, the single neuron, and multi-dimensional stimuli

Abstract

Sensory processing disorders are characterized by poor behavioral performance in psychophysical tasks relative to the general population. To assess the etiology of those deficits, a tempting possibility is to look for abnormalities in the information available in responses of specific groups of neurons, which could then become the target of rehabilitation. I discuss here basic limitations in linking behavioral performance to neuronal responses. Our understanding of sensory neurons comes from measuring tuning curves, which describe how the response varies when a dimension of the stimulus is varied. In reality, however, neural responses are modulated by many stimulus features, e.g., in the auditory system: amplitude and spectrum of the stimulus, spatial position, temporal properties, etc. I illustrate the challenges that this tuning heterogeneity raises when trying to predict results of behavioral experiments from neuronal responses. I first exhibit effects in theory using Fisher Information, then analyze explicit optimal decoders trained to estimate spectral and spatial properties of an auditory stimulus from neural responses. I show how heterogeneity in tuning is a major contributor to the robustness of the encoding of multiple stimulus dimensions. Finally, I described how this changes the interpretation of neural constraints on behavioral acuity.