Efficient neural population codes

The brain faces at least two challenges critical to an animal\u27s survival: to encode sensory stimuli generated by the physical world, and to produce behavior which is well-adapted to the current external context. The problem is further complicated by the fact that the animal is acting within a dynamic environment; many neural computations must therefore be flexible and adaptive. Furthermore, there is clear survival value in having the computations be efficient: evolution should favor neural coding schemes that make best use of the available resources. This dissertation will present new evidence for efficient coding in human visual perception, theoretical predictions of optimal network architecture in a population of sensory neurons, and experimental tests of the predictions of the efficient coding principle applied to the retinal representation of a changing visual environment. To study some of these questions experimentally, it was necessary to develop novel methods for analyzing experimental data taken from many neurons simultaneously. These methods will also be described. Finally, the dissertation will address the question of how neural populations can support optimal behavior by presenting a theoretical analysis connecting the brain\u27s representation of space to the solution of an optimal path-planning problem