Partial recognition: The P3 marks the top-down similarity between task-relevant targets and presented stimuli

The P3 is a psychophysiological marker of visual recognition, being related to object detection, memory encoding, and action selection. Yet computing the similarity of a presented stimulus to a represented target is an ill-defined task for stimuli such as faces, given the plenitude of potentially relevant features. We previously proposed that current neural networks can define stimuli both at their objective, physical and subjective, cognitive levels, and thus compute similarity. This similarity - or perception/representation distance inverse - linearly predicted P3 amplitude in a face identity recognition task. However, this left open whether the distance-P3 relation reflected top-down task-related recognition or bottom-up family-resemblance repetition effects. We therefore revisited the paradigm but included precisely matched presentations of foils: task-irrelevant images sampled and presented at target-matched distances. The results showed that an early binary differentiation between targets and other images occurred at the N170 latency while for the P3, a clear effect of distance was found: the larger the distance, the smaller the P3. The effects of target-relevance were clearly dissociable from the effect of perceiving foils, which did not affect the N170 at all, and showed only a minor, binary effect on the P3, unrelated to the specific distance. Taken together, we argue that keeping a target in visual working memory involves an early, top-down mechanism, which evaluates the evidence for making a perceptual decision. This mechanism is followed soon after by a more passive, bottom-up process, which updates both the probability and representation of stimulus identities.