January 24th14:30, room 2014 (Shannon 660) (see location):
Daniela Pamplona (Biovision team, INRIA Sophia-Antipolis / TAO)
Title: Data Based Approaches in Retinal Models and Analysis
Abstract:In this talk I will speak about data-based approaches to understand information processing in the retina at two levels. In the first part, I will show how a detailed model of the input to the visual system lead to a efficient coding explanation of the retinal variability across the visual field. On the second part I will introduce the discrete leak integrate and fire network (DLIF-N) and some properties of this model. Thereafter, I consider the inverse problem: “Can we estimate the underlying network from the data?” Considering the Kullback Leibler Divergence minimization method, I show that “yes, we can!”, but the solution is not unique. Interestingly, with the learned connectivity we can reproduce low and high order statistics of the ground truth data. In the end I will discuss ongoing work and possible future directions.
Contact: guillaume.charpiat at inria.fr