Co-occurrence of past and present shifts current neural representations and mediates serial biases

Abstract

The regularities of the world render an intricate interplay between past and present. Even across independent trials, current-trial perception can be automatically shifted by preceding trials, namely the 'serial bias'. Meanwhile, the neural implementation of the spontaneous shift of present by past that operates on multiple features remains unknown. In two auditory categorization experiments with human electrophysiology recordings, we demonstrate that serial bias arises from the co-occurrence of past-trial neural reactivation and the neural encoding of current-trial features. The meeting of past and present shifts the neural representation of current-trial features and modulates serial bias behavior. Critically, past-trial features (i.e., pitch, category, motor response) that constitute an 'event-file' keep their respective identities in working memory and are only reactivated by the corresponding features in the current trial, giving rise to dissociated feature-specific serial biases. This 'event-file' reactivation might constitute a fundamental mechanism for adaptive past-to-present generalizations over multiple features.

Distinguishing response from stimulus driven history biases

“…We further found evidence for this attraction extending back 6 trials and separate evidence for a repulsion from the physical identity of the stimulus for trials 2, 3, 5 and 6 trials back. This pattern matches prior observations and supports the idea that the stimulus presentation leads to a repulsive bias at encoding while more high-level decisional representations impose a prior of stability (Braun et al 2018; Moon and Kwon 2022; Papadimitriou, White, and Snyder 2016; Pascucci et al 2019; Pegors et al 2015; Sadil et al 2021; Sheehan and Serences 2022; Zhang and Alais 2020; Zhang and Luo 2022). Such a framework additionally fits with general frameworks like efficient encoding and Bayesian inference seen in perception (Wei and Stocker 2015) and pattern separation and completion seen in various networks across the brain (Cayco-Gajic and Silver 2019).…”

“…Now that there are several studies showing strong evidence for response over stimulus driven effects (Moon and Kwon 2022; Sadil et al 2021), the goalposts have shifted to further disambiguate exactly which response related components are driving these effects. Change detection paradigms or generally un-correlating responses from perception offer promising avenues to explore this possibility further (Braun, Urai, and Donner 2018; Sheehan et al 2022; Zhang and Luo 2022). That said, we argue here that examining biases just as a function of the physical identity of the previous stimulus is ignoring the important role of other biases in shaping the perception of current and past stimuli and may lead to an under and mismatched measurement of the true underlying bias (Pascucci et al 2019; Sadil et al 2021).…”

Reactivated past decisions repel early sensory processing and attract late decision-making

“…PFC recordings on monkeys show that past-trial information is transiently reactivated just before trial onset and further correlates with serial bias behavior 28 . Our recent study 19 as well as previous works [31][32][33][34] demonstrate that past-trial information, indeed retained in activity-silent states during the intertrial interval, is triggered by the corresponding event within the current trial, and the co-occurrence of past and present contributes to serial bias. Most importantly, the past-to-present neural influence exhibits a feature-specific direction commensurate with the corresponding bias behavior, i.e., repulsive for sensory and motor features, and attractive for choices 19 .…”

“…Our recent study 19 as well as previous works [31][32][33][34] demonstrate that past-trial information, indeed retained in activity-silent states during the intertrial interval, is triggered by the corresponding event within the current trial, and the co-occurrence of past and present contributes to serial bias. Most importantly, the past-to-present neural influence exhibits a feature-specific direction commensurate with the corresponding bias behavior, i.e., repulsive for sensory and motor features, and attractive for choices 19 . Serial bias is also posited to depend on neural representational similarities between sensory and memory information 35 , such as aligned and flipped relationships resulting in attractive and repulsive serial bias 26,33 .…”

“…One striking example is serial dependence, that is, perception tends to be involuntarily biased towards stimuli in previous trials [4][5][6][7][8] . Serial dependence has been robustly observed in a wide range of visual features, from simple ones like orientation 4,[9][10][11][12] , numerosity 13 , and position 14,15 to abstract ones such as facial attractiveness 16,17 and other modalities 18,19 . Despite its detrimental effects on perceptual fidelity, serial dependence is thought to reflect an optimizing strategy of the brain to increase perceptual stability and efficiency by taking advantage of inherent temporal correlations in a world that is presumably stable over a short time scale 20,21 .…”