Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session C: Friday, August 15, 2:00 – 5:00 pm, de Brug & E‑Hall

Bistable perception emerges from loopy inference in strongly coupled probabilistic graphs

Alexandre Garcia-Duran¹, Martijn Wokke, Manuel Molano-Mazon², Alexandre Hyafil¹; ¹Centre de Recerca Matemàtica, ²Universidad Politécnica de Cataluna

Presenter: Alexandre Garcia-Duran

During perception, the brain continuously processes sensory input, selecting among competing interpretations and assigning certainty -confidence- to each. Typically, confidence correlates with the strength of sensory evidence. In bistable perception, however, one interpretation is confidently perceived at a time, yet perception alternates despite no changes in the stimulus. We investigate which properties of visual stimuli drive this dissociation between evidence strength and confidence. We propose that bistability arises from approximate probabilistic inference over an internal representation of a stimulus with strongly coupled features. Using the Necker cube as an example, we model how perceived depth at each vertex is coupled with its neighbors, reflecting natural co-occurrence statistics. We analyze the dynamics of three inference algorithms. In all cases, strong feature coupling introduces loops in the internal representation that stabilize one percept, while internal noise drives perceptual switches. This creates a double-well potential, with perception fluctuating between high-confidence states. To test this, we designed a bistable stimulus in which feature coupling and sensory strength were independently manipulated. Our results show that stronger coupling leads to higher reported confidence, even when sensory evidence is weak. These findings suggest that bistable perception results from internal inference dynamics when stimulus features are tightly coupled.

Topic Area: Predictive Processing & Cognitive Control

Extended Abstract: Full Text PDF