Independent history traces for distinct percepts derived from a single stimulus
Toso, A.; Diamond, M. E.
Show abstract
Perception is more than a direct readout of current sensory input; it is also shaped by recent experience, a phenomenon known as serial dependence. While the effects of recent experience have been well characterized for single stimulus features, it remains unclear how perceptual history operates when multiple perceptual dimensions must be extracted from a single sensory event. Here, human participants reported their judgment of vibrotactile stimuli, with the relevant feature - either the intensity or duration - randomly cued on each trial. Within trials, intensity and duration interacted bidirectionally - longer stimuli were perceived as stronger and stronger stimuli as longer - indicating that both features are derived from a shared sensory representation. Across trials, however, serial dependence was feature-specific: perceived intensity was selectively attracted toward previously perceived intensities, and perceived duration toward previously perceived durations. The current trial judgment was influenced by perceptual estimates from multiple preceding trials, but only those in which the same feature was task-relevant. We formalize these findings with a computational model in which perceptual history is maintained in parallel memory traces, each selectively updated when the corresponding feature is task-relevant. The model captured both the psychometric structure of behavior and the feature-specific, multi-trial dependence of perceptual judgments, with traces that accumulate past percepts and bias current estimates along the corresponding feature dimension.Serial dependence in this task operates on perceptual representations downstream of a shared sensory encoding stage, and is maintained simultaneously in feature-specific, task-gated memory traces within higher-level cortical circuits.
Matching journals
The top 3 journals account for 50% of the predicted probability mass.
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.