József Fiser

15 février 2013

Séminaire Tutoré

Department of Cognitive Science, Central European University, Budapest, Hungary

A sampling-based framework for probabilistic perceptual representation and computation in the cortex

invité par Frédéric Chavane


bstract : The mind’s internal representations of its environment have a crucial role in the emergence of intelligent behavior, yet there are few concrete proposals about the nature of these internal representations or the way they are acquired. Using the domain of visual recognition, I will present a framework and a combined empirical-computational program that explore these issues. I will start from the point that in everyday perceptual tasks humans and animals process not only the sensory information but also their uncertainty about that information, and they do this in a theoretically optimal probabilistic manner. I will focus on the issue of whether the implementation of such representations and computations are biologically and computationally feasible in the brain. I will briefly outline how probabilistic internal representations could be implemented in the cortex in a sampling-based manner, and how this can explain a wide range of puzzling observations such as illusions and dreams, as well as the recorded trial-to-trial variability, high level of spontaneous activity in the brain. I provide a first confirmation of this framework by demonstrating that as young animals grow, the visually evoked and spontaneous activity in their brains becomes statistically similar, indicating how their internal model gets tuned to the structure of their environment. Next I will cover some or all of the following issues as time permits : a) what additional evidence we found that those developmental changes supporting the framework are related to learning, b) how the probabilistic optimal interpretation compares to alternative explanations, c) how computationally feasible this framework is, d) what unexplained physiological data in the literature the sampling framework can explain, e) how low in visual processing one can go behaviorally to show evidence of sampling based computation in the brain, f) how the framework can be extended to decision making and attention, including comparisons to traditional models (integration-to-bound, etc.). I will round-up by presenting the new challenges this framework faces. Finally, I will conclude that the sampling-based probabilistic framework offers a rigorous approach to exploring empirically the structure of internal representations in the brain and it might provide a surprisingly comprehensive answer to the age-old question of body-mind duality.

CNRS logo université Aix Marseille logo | plan du site | mentions légales | contact | admin | intranet | intcloud |