Bayesian Comparison of Explicit and Implicit Causal Inference Strategies in Multisensory Heading Perception

Abstract

The precision of multisensory heading perception improves when visual and vestibular cues arising from the same cause, namely motion of the observer through a stationary environment, are integrated. Thus, in order to determine how the cues should be processed, the brain must infer the causal relationship underlying the multisensory cues.In heading perception, however, it is unclear whether observers follow the Bayesian strategy, a simpler non-Bayesian heuristic, or even perform causal inference at all. We developed an efficient and robust computational framework to perform Bayesian model comparison of causal inference strategies, which incorporates a number of alternative assumptions about the observers. With this framework, we investigated whether human observers' performance in an explicit cause attribution and an implicit heading discrimination task can be modeled as a causal inference process. In the explicit inference task, all subjects accounted for cue disparity when reporting judgments of common cause, although not necessarily all in a Bayesian fashion. By contrast, but in agreement with previous findings, data from the heading discrimination task only could not rule out that several of the same observers were adopting a forced-fusion strategy, whereby cues are integrated regardless of disparity. Only when we combined evidence from both tasks we were able to rule out forced-fusion in the heading discrimination task. Crucially, findings were robust across a number of variants of models and analyses. Our results demonstrate that our proposed computational framework allows researchers to ask complex questions within a rigorous Bayesian framework that accounts for parameter and model uncertainty.

The role of sensory uncertainty in simple contour integration

“…In such tasks, the 293 observer receives sensory measurements of possibly discrepant cues from distinct sensory 294 modalities (e.g., vision and hearing), and has to infer whether the cues originated from 295 the same source (C = 1) or from different sources (C = 0) -leading to, respectively, cue 296 integration and cue segregation. Previous work has shown that Bayesian causal inference 297 models provide a good qualitative description of human performance in multisensory 298 perception with discrepant cues, but quantitative comparison hints at deviations from 299 exact Bayesian behavior [30], not unlike what we find here. Our study differs from 300 previous work in that here we focus on an atomic form of perceptual organization.…”

“…25 A fully Bayesian approach to perceptual organization would provide a normative 26 way for dealing both with high-level uncertainty arising from ambiguity in the latent 27 structure of the scene, and with low-level (sensory) uncertainty arising from noise in 28 measuring primitive elements of the scene. Crucially, however, previous studies in 29 perceptual organization have not examined whether the decision rule adapts flexibly as 30 function of sensory uncertainty. Such adaptation is a form of probabilistic computation 31 and is one of the basic signatures of Bayesian optimality [7].…”

“…http://dx.doi.org/10.1101/350082 doi: bioRxiv preprint first posted online Jun. 18, 2018; that is an estimate of the intrinsic variability of the data which represents an upper 262 bound on the performance of any model [30] …”

“…We fitted the models by drawing samples from the unnormalized log posterior 438 distribution of the parameters p(θ|data) using Markov Chain Monte Carlo (parallel slice 439 sampling [30,45]) for each subject. The posterior distribution of the parameters is 440 proportional to the sum of data likelihood (Eq 3) and a factorized prior over the each 441 parameter j, 442 log p(θ|data, model) = log p(data|θ, model) + j log p(θ j |model) + const.…”

Further Perceptions of Probability: In Defence of Associative Models – A Commentary on Gallistel et al. 2014

“…Therefore, 532 we use a simplified, "custom" likelihood function for model fitting (Appendix B). We use the 533 Bayesian Adaptive Direct Search (BADS) method (Acerbi & Ma, 2017) to find the parameters 534 that maximise this function. In order to reduce the risk of terminating in local maxima, we run 535 BADS thirty times with different initial parameter values.…”