Context-dependent sensory adaptation in cortical area MT as a substrate of flexible decision-making

Visual decisions typically require accumulation of uncertain sensory information to make inferences about the state of the world. This inference process is complicated by the fact that the state of the world can change, which often requires adjustments in how sensory information is accumulated. Previous work has modeled optimal inference in these changing environments [1-2]. One key feature of these models is a leakage of accumulated evidence, where the leak rate depends on environmental stability. For example, in more unstable environments, an increased leak rate helps remove pre-change evidence and thus attenuate beliefs that would hinder choice accuracy. This kind of context-dependent leak can describe human decision-making behavior, but its neural implementation remains unknown. We propose that this leak involves, at least in part, context-dependent changes in cortical sensory adaptation during evidence encoding. Specifically, we hypothesize that the adaptation dynamics of evidence encoding are modulated by the rate of change of recent sensory inputs, enabling the decision process to adjust to stimulus statistics.

We trained two macaques on a random-dot motion task where we manipulated context stability via “change-points,” or abrupt switches in motion direction. We recorded activity of neurons in middle temporal area (MT), which contribute causally to decisions about motion direction [3]. Preliminary results suggest that monkeys were less sensitive to evidence as a function of viewing time in more unstable contexts (i.e., more change-points), which is consistent with a leak that depends on stimulus context. Moreover, neural responses in MT were more strongly adapted in an unstable versus stable context. This context-dependent adaptation both diminished the ability of single neurons to discriminate between motion directions and related to the monkey’s behavioral sensitivity to evidence during the task. Collectively, these results suggest that recent stimulus dynamics can induce changes in evidence encoding that contribute to flexible perceptual decisions.