Event:

With the recent emergence of new techniques for simultaneous recordings from large populations of neurons in behaving animals, it is becoming particularly critical that we understand how neural circuits compute with population codes and how these computations lead to behavior. Most theories of neural coding treat the nervous system as a noisy device and seek neural codes that are robust to the presence of this internal noise. In sharp contrast, we argue that the limiting factor on behavioral performance, and more generally on coding and computation, is not internal noise but the limited information provided by the world and our sensors as well as the approximate computations performed by neural circuits. We will demonstrate this idea with a simple model of orientation selectivity. We will show that this model predicts the existence of a particular pattern of information-limiting correlations in population codes which we call differential correlations. It also predicts a specific relationship between the discrimination threshold of individual neurons and their correlations with the response of the animal. We will present experimental evidence in support of both predictions.