Abstract
Competing models of decision-making invoke distinct mechanisms underlying the integration of sensory evidence. Prior work has used stimulus and response timing to probe this mechanism. Here, we adopt a complimentary strategy in which we psychophysically measure and manipulate the time course of strategic weighting of sensory information used to inform decisions. Observers performed a motion discrimination task in which each trial comprised seven discrete epochs in time (i.e. seven "motion pulses") where each can have distinct motion coherence. On each trial, the observer reported net leftward or rightward perceived motion. Trials contained varied motion strength, with a subset containing purely directional noise. These noise trials were used for reverse correlation analysis, in which the choice was correlated with the motion strength in each pulse, yielding a temporal kernel describing the observer's temporal weighting strategy. Baseline kernels varied across both human and monkey observers, but on average demonstrated preferential weighting of early pulses, as often observed in other work. After measuring an observer's baseline temporal weighting, we attempted to change their strategy by modifying the likelihood of high coherence at different times during non-noise trials. By biasing strong motion to occur late in the stimulus, we were able to shift early weighting to late weighting— thus demonstrating direct control over the decision strategy. Although the time course of stimulus weighting can be taken as evidence for a particular type of integration mechanism, many models of decision-making are often flexible enough to account for many patterns of temporal stimulus weighting. Our ability to manipulate the time course of weighting suggests that it should be thought of as a malleable strategy that observers can flexibly adapt to stimulus statistics. This approach will be used in ongoing electrophysiological experiments as a causal manipulation to probe the temporal structure of choice-correlated neuronal activity.
Meeting abstract presented at VSS 2016