A Bayesian and efficient observer model explains concurrent attractive and repulsive history biases in visual perception

Abstract

Perceptual decisions can be repelled away from (repulsive adaptation) or attracted towards recent visual experience (attractive serial dependence). It is currently unclear whether and how these repulsive and attractive biases interact during visual processing and what computational principles may underlie these history dependencies. In the current study, we disentangle repulsive and attractive biases by exploring the respective timescales over which current visual processing is influenced by previous experience. Across four experiments, we find that perceptual decisions about stimulus orientation are concurrently attracted towards the short-term perceptual history and repelled from stimuli experienced up to minutes into the past. We show that the temporal pattern of short-term attraction and long-term repulsion cannot be captured by an ideal Bayesian observer model alone. Instead, it is well captured by an ideal observer model with efficient encoding and Bayesian decoding of visual information in a slowly changing environment. Concurrent attractive and repulsive history biases in perceptual decisions may thus be the consequence of the need for visual processing to simultaneously satisfy constraints of both efficiency and stability.

Expectations about motion direction affect perception and anticipatory smooth pursuit differently

“…In a speed estimation task in which observers had to judge whether the speed in the current trial was faster or slower than the average speed across all previous trials, observers tended to overestimate a current target’s velocity when the average velocity across previous trials was slow and vice versa for fast velocity (Maus et al 2015). Similar repulsion biases have been found in studies in which observers had to adjust the orientation of a test stimulus relative to an inducer stimulus, when both stimulus orientations differed by more than 60° between the previous and current trial (Fritsche et al 2020). In this scenario, observers’ adjustment responses were sometimes repelled away from the previous trial’s stimulus orientation.…”

“…We do not know how robust our results are with regard to parametric variations of the visual stimuli. For example, an attraction bias is mostly observed in studies using moving stimuli whose directions differed by about 60° or less (Fritsche et al 2020) rather than 180° (as in our study). In addition, effects of expectation are often examined with weak motion stimuli, i.e., low contrast (Chalk et al 2010) or low coherence (probe trials in our study), because the Bayesian integration hypothesis postulates that the effect of expectation would be larger on a stimulus with less reliable sensory signals.…”

“…However, there is evidence that modulation in early sensory cortices, from primary visual cortex (V1) to MT and MST, might underlie repulsion and attraction biases in perception. For example, the repulsion bias in perceived orientation was found to be stronger when the current and previous stimuli in a given trial were presented at the same location (Fritsche et al 2020). These findings indicate that this orientation bias was driven by effects that are spatially specific (and retinotopically congruent), which likely implies modulation of neurons in early sensory cortex, such as V1, responsive to stimuli within small receptive fields.…”

“…Anticipatory pursuit and perception rely on history on different temporal scales (Maus et al 2015). Attraction and repulsion perceptual bias seem to also operate on different time scales (Chopin and Mamassian 2012; Fritsche et al 2020), and could occur with slight changes of parameters in the same paradigm (Kanai and Verstraten 2005). Therefore, understanding the temporal dependency and development of the effects of expectation would be crucial to understand the complicated interaction between attraction and repulsion biases.…”

Temporal dynamics of implicit memory underlying serial dependence

“…How long does the serial dependence effect persist between trials? Previous studies have shown that the impact of a trial on future behavior decreases with time (Fischer and Whitney, 2014; Kalm and Norris 2018; Fritsche et al 2020). However, this decrease in serial dependence might be explained by an overwriting of the representation of previously experienced stimuli or responses (Matthey, Bays and Dayan 2015) or by decreasing their representation precision due to interference (Kalm and Norris 2018), not temporal decay alone.…”