Allocation of Computational Resources in the Nervous System.

The nervous system integrates past information together with predictions about the future in order to produce rewarding actions for the organism. This dissertation focuses on the resources underlying these computations, and the task-dependent allocation of these resources. We present evidence that principles from optimal coding and optimal estimation account for overt and covert orienting phenomena, as observed from both behavioral experiments and neuronal recordings. First, we review behavioral measurements related to selective attention and discuss models that account for these data. We show that reallocation of resources emerges as a natural property of systems that encode their inputs efficiently under non-uniform constraints. We continue by discussing the attentional modulation of neuronal activity, and showthat: (1) Modulation of coding strategies does not require special mechanisms: it is possible to obtain dramatic modulation even when signals informing the system about fidelity requirements enter the system in a fashion indistinguishable from sensory signals. (2) Optimal coding under non-uniform fidelity requirements is sufficient to account for the firing rate modulation observed during selective attention experiments. (3) The response of a single neuron cannot bewell characterized by measurements of attentional modulation of only a single sensory stimulus. (4) The magnitude of the activity modulation depends on the capacity of the neural circuit. A later chapter discusses the neural mechanisms for resource allocation, and the relation between attentional mechanisms and receptive field formation. The remainder of the dissertation focuses on overt orienting phenomena and active perception. We present a theoretical analysis of the allocation of resources during state estimation of multiple targets with different uncertainties, together with eye-tracking experiments that confirm our predictions. We finish by discussing the implications of these results to our current understanding of orienting phenomena and the neural code