VSS 2023 - A Bayesian efficient observer model to explain attractive and repulsive temporal context effects when perceiving multistable dot lattices


In multistable dot lattices, the orientation we perceive is attracted towards the orientation we perceived in the immediately preceding stimulus and repelled from the orientation for which most evidence was present previously (Van Geert et al., 2022). Theoretically-inspired models have been proposed to explain the co-occurrence of attractive and repulsive context effects in multistable dot lattice tasks, but these models artificially induced a direct influence of the previous trial on the current (Gepshtein & Kubovy, 2005; Schwiedrzik et al., 2014). We conducted a simulation study to test whether the observed attractive and repulsive context effects could be explained by a Bayesian efficient observer model formulated by Wei and Stocker (2015), which has previously been applied successfully to different tasks involving non-ambiguous stimuli (e.g., Fritsche et al., 2020, Langlois et al., 2021). The Bayesian efficient observer model assumes variable encoding precision of orientations in line with their frequency of occurrence (i.e., efficient encoding) and takes the dissimilarity between stimulus space and sensory space into account, which leads to different outcome predictions depending on levels of external stimulus and internal sensory noise. Although a first, more direct implementation of the Bayesian efficient observer model could not predict the empirically observed attractive effect of the previous percept, a slightly adapted version of the model including a direct influence of the previous percept on the prior for the current percept could explain the co-occurrence of both temporal context effects. To conclude, this simulation study provides evidence that information on the previous percept is needed on top of the previous stimulus evidence to adequately model both attractive and repulsive context effects when perceiving multistable dot lattices. This conclusion is in line with Mao and Stocker (2022), who suggested to take higher-level representations into account to adequately model human subjects’ orientation percepts.

May 23, 2023 08:30
Banyan Breezeway, TradeWinds Island Grand, St. Pete Beach, Florida, USA