Perception is often described as a predictive process based on an optimal inference with respect to a generative model. We study here the principled construction of a generative model specifically crafted to probe motion perception. In that context, we first provide an axiomatic, biologically-driven derivation of the model. This model synthesizes random dynamic textures which are defined by stationary Gaussian distributions obtained by the random aggregation of warped patterns. Importantly, we show that this model can equivalently be described as a stochastic partial differential equation. Using this characterization of motion in images, it allows us to recast motion-energy models into a principled Bayesian inference framework. Finally, we apply these textures in order to psychophysically probe speed perception in humans. In this framework, while the likelihood is derived from the generative model, the prior is estimated from the observed results and accounts for the perceptual bias in a principled fashion.
Jonathan Vacher (Université Paris Dauphine)
Andrew Isaac Meso (Institut des neurosciences de la Timone)
Laurent U Perrinet (Institut des neurosciences de la Timone)
Laurent Perrinet is a computational neuroscientist specialized in large scale neural network models of low-level vision, perception and action, currently at the "Institut de Neurosciences de la Timone" (France), a joint research unit (CNRS / Aix-Marseille Université). He co-authored more than 40 articles in computational neuroscience and computer vision. He graduated from the aeronautics engineering school SUPAERO, in Toulouse (France) with a signal processing and applied mathematics degree. He received a PhD in Cognitive Science in 2003 on the mathematical analysis of temporal spike coding of images by using a multi-scale and adaptive representation of natural scenes. His research program is focusing in bridging the complex dynamics of realistic, large-scale models of spiking neurons with functional models of low-level vision. In particular, as part of the FACETS and BrainScaleS consortia, he has developed experimental protocols in collaboration with neurophysiologists to characterize the response of population of neurons. Recently, he extended models of visual processing in the framework of predictive processing in collaboration with the team of Karl Friston at the University College of London. This method aims at characterizing the processing of dynamical flow of information as an active inference process. His current challenge within the <a href="https://www.int.univ-amu.fr/spip.php?page=equipe&equipe=NeOpTo&lang=en">NeOpTo team</a> is to translate, or *compile* in computer terminology, this mathematical formalism with the event-based nature of neural information with the aim of pushing forward the frontiers of Artificial Intelligence systems.
Gabriel Peyré (CNRS and Ceremade, Université Paris-Dauphine)
More from the Same Authors
2015 Poster: Biologically Inspired Dynamic Textures for Probing Motion Perception »
Jonathan Vacher · Andrew Isaac Meso · Laurent U Perrinet · Gabriel Peyré