On the inseparability of the prior and neural resources in behavioural bias

Abstract

Sensory decision making depends on environmental expectations and finite neural resources, yet these quantities are often modelled as separable values in biological systems. Here we show that this separation is neither computationally nor biologically necessary for efficient population codes. We demonstrate that prior expectations can be embedded directly in an encoding populations tuning, thereby producing behavioural biases via Bayes-optimal inference with a uniform explicit prior. Re-analysis of V1 physiology provides preliminary support for the predicted organisation of population activity. These results highlight how systematic perceptual bias can stem from the implementation of decoding in an efficiently organised sensory code.