Jaime de la Rocha
Date: April 04, 2017. 12:00
Location: CCU Seminar Room
title: Integration of information in perceptual decisions across different time scales.
Affiliation: Institut d’Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS), Barcelona.
Animal perceptual decisions not only reflect current sensory information integrated over short time scales (e.g. 1 s) but are also shaped by recent experience. Although there exist mechanistic models of stimulus integration, there is still little understanding about the mechanisms underlying history-dependent decision biases. In the first part of this talk I will present data from a novel two-alternative forced choice auditory discrimination task in rats, in which the probability to repeat the previous stimulus category was varied in blocks of trials. Rats adapted to this environment by developing a strategy that capitalized on the serial correlations of the stimulus sequence: a bias towards repeating the same response built up after correct repetitions, and conversely an alternation bias developed after correct alternations. We quantified this strategy using a Generalized Linear Model (GLM) and found that the choice bias could be flexibly and transiently put aside after error choices when possibly the reliability of the internal model was questioned and recovered once the animal made again a correct choice. In the second part of the talk, I will present an analysis of the dynamics of stimulus integration in models with attractor dynamics during a fixed duration 2AFC tasks. We studied a spiking network and a diffusion process in a double well (DW) potential (the one dimensional potential equivalent to the attractor spiking network). We considered a fluctuating stimulus in which the evidence is drawn from a distribution with mean ? and variance ?. The input fluctuations plus some intrinsic internal noise caused stochastic transitions between decision attractors that can be correcting transitions if the final well is deeper than the initial or error transitions in the opposite case. Analytical and numerical analysis show that (1) as ? increased the stimulus fluctuations with a higher impact on choice shifted from the beginning (transient integration also called primacy) to the end of the stimulus (leaky integration or recency). This occurred because for small ?, after reaching an attractor the system was unlikely to escape, thus the first stimulus fluctuations causing the first attractor visit had highest impact on choice. When ? was large, multiple attractor transitions occurred, and were the late stimulus fluctuations causing the last transition the ones that determined the final decision. (2) The categorization performance showed a resonance with a local maximum at ? max > 0. I will finally present some preliminary data from a perceptual categorization experiment in humans and rats that try to test these predictions.