Random stimuli, perceptual scales, and the probabilistic brain

Abstract: 

Bayesian models of perception are appealing, influential, and often surprisingly successful. But they are also frequently tested with stimuli that may be too simple to really challenge them. In this seminar, I will revisit this issue through psychophysical experiments based on random visual stimuli.

I will start from the efficient-coding framework of Wei and Stocker and its recent extensions by Hahn and Wei, and argue that an important tension appears when stimulus uncertainty is manipulated experimentally. If perceptual encoding reflects long-term environmental statistics, should it really change with every change in stimulus noise introduced in the lab? This question motivates a fuller Bayesian treatment of perceptual tasks.

I will then present recent results in vision using orientation and spatial-frequency stimuli in both 2AFC and MLDS. Looking at these two paradigms side by side reveals that they do not simply tell the same story, and raises a broader question: when a stimulus changes both in mean and in spread, what is the observer actually representing?

This opens the door to a more general discussion. Perhaps the relevant perceptual object is not always a single feature value, but sometimes something richer. I will illustrate this with stimuli in which several aspects vary at once, and use this to question what we really mean by the “variable of interest” in perception.

I will finish with an opening toward motion perception, in vision and audition. In particular, I will discuss how continuous random-motion stimuli might help design cleaner experiments for auditory motion perception on a loudspeaker array, avoiding some classic confounds tied to motion onset and offset.

‼️ Le séminaire peut être suivi en ligne via le lien : https://univ-amu-fr.zoom.us/j/95011527683?pwd=bTeNvaaZY75buaa7bRL0pcucXSASm7.1

Le mardi 23 juin 2026 à 11h00 / Amphithéâtre François Canac, LMA

Voir en ligne : plus d’informations concernant l’orateur