Behavioral and neural population dynamics of foraging decisions

Abstract

The adaptive value of the brain lies in its ability to generate a complex behavioral repertoire. This ability arises out of the computations that occur as sensory input is transformed into motor output. The process of behavioral selection, or decision making, as a collection of neurophysiological algorithms is therefore a useful paradigm for understanding how the brain works. Here, I argue for an approach to characterizing these algorithms that combines behavioral and theoretical insights with neurophysiology itself. I do so via three studies of decision making that highlight this confluence. In macaques performing a computerized foraging task, I observe that pupillary responses to potential reward outcomes appear to reflect relative (rather than absolute) reward value. The correlation of pupil size with reward also reverses once a decision has been made, suggesting a dynamic underlying computation. Next, I hypothesize that a widely observed suboptimal foraging behavior could be theoretically explained by accounting for the variability intrinsic to such dynamic decision making processes. Using a mathematical model of a patch foraging task, I show that representing accumulated reward thresholds as noisy distributions instead of specific values shifts the optimal strategy towards what is behaviorally observed. Finally, in order to characterize what a decision making algorithm based on relative value might physiologically entail, I analyze population dynamics in the orbitofrontal cortex. I find that higher offer values during the second epoch of a sequential decision making task elicit stronger perturbations of the population state. The strength of this perturbation, however, is biased by the population state entropy at the beginning of the epoch. The entropy, in turn, is systematically related to the value of the first offer, providing a potential algorithm for relative value comparison.