Temporal integration of feature probability distributions in visual working memory

Visual memory has remarkable effects on human search behavior. In particular, when target features are repeated, search efficiency increases. Similar effects are also found for repeating distractors. Conversely, when a distractor feature and target feature reverse their roles, search times are slowed down. Recent studies have revealed that the visual system is not only sensitive to distractor features per se, but the actual distractor feature probabilities. Changes in search times were determined not only by whether that particular feature characteristic was a distractor but also the frequency of that distractor feature over consecutive trials: Most probable distractors produced the strongest role reversal, while less probable distractors produced weaker role reversals. These search displays involved many distractor exemplars on each trial, but whether observers can learn distributions where only a single exemplar from a distribution is presented on each trial remains unknown. Here, we investigated whether target probability distributions can be encoded in working memory. Over blocks of trials (144 trials per block) observers searched for an oddly-colored target that was drawn from either a Gaussian or uniform distribution. Not only was search influenced by the repetition of a target feature but more interestingly also by the probability of that feature within the block. The same targets, coming from the extremes of the two distributions were found significantly slower when distractors were drawn from a Gaussian distribution than from a uniform distribution indicating that observers were sensitive to the target probability. In a subsequent experiment we replicated the effect using binned distributions and moreover discovered the limitations of target distribution encoding by using bimodal target distributions. Our results demonstrate detailed internal representations of target probability distributions in working memory and the visual system’s ability to integrate single exemplars into probability distributions over surprisingly long trial sequences.