Representational capacity of neural codes in the cortex

The coding of sensory stimuli in the neural response is a fundamental property of neural systems that impacts many of its properties. The coding scheme also determines how many different stimuli a population of neurons can represent. Although the coding of single stimuli has been studied extensively, coding of multiple stimuli has been studied far less. Here we study the representational capacity in the visual system when stimulus pairs are represented simultaneously. We assume that the response to the individual stimuli is given, and that the neurons interacts non-linearly to form the response to the pair. Using a Bayesian read-out we find that using a linear sum leads to a smaller capacity, than a maximum-like interaction. Thus non-linear interaction improve coding capacity. The results provide a novel interpretation for the non-linear interaction observed experimentally.