|
Sensory response variability: biasing or reflecting perceptual decisions?| old_uid | 10110 |
|---|
| title | Sensory response variability: biasing or reflecting perceptual decisions? |
|---|
| start_date | 2011/07/04 |
|---|
| schedule | 15h-16h |
|---|
| online | no |
|---|
| location_info | salle de réunion du laboratoire |
|---|
| summary | Trial-to-trial variability in the response of cortical sensory neurons to ambiguous stimuli correlates with variability in the perception. This correlation, usually quantified by the Choice Probability (CP), has been interpreted to reflect the bottom-up causal role of neuronal variability on guiding perceptual decisions[1] although recent evidence suggests that the time course of CP is partially inconsistent with this interpretation [2]. We study two network mechanisms that generate CP in a network model composed of a sensory circuit reciprocally connected with a decision circuit: (i) correlated spiking variability in the sensory population and (ii) top-down inputs from the decision circuit. The model circuits are composed of two recurrently connected excitatory populations with opposite stimulus preference (e.g. UP vs. DOWN motion) and an inhibitory population. Parameters were chosen so that the sensory circuit showed approximately
linear behavior and the decision circuit winner-take-all dynamics [3]. We
first investigated a low dimensional description of the network using coupled rate equations. Each rate variable represents the activity of one population and receives an independent noise term, whose magnitude captures ad-hoc the degree of spiking correlation among cells in each population. We compare two versions of this network: (i) feedforward model (FF), with large spiking correlations in the sensory populations and no top-down inputs; (ii) feedback model (FB), with small spiking correlations and top-down inputs. We computed the trial-averaged responses of one sensory population conditioned on the two possible decisions and took their difference as a proxy of CP. The time course of this difference in the FF model peaked right after stimulus onset and then decayed slowly to zero whereas in the FB model it rose slowly and reached a plateau. Thus, a causal role of correlated spiking variability would be reflected in the early rise and decay of the CP, while slow CP saturation mirrors the arrival of the decision circuit to the final choice attractor. In the FB model, the top-down input promotes perceptual stability because it amplifies small initial differences in the response of the sensory populations, which are then sent bottom-up enhancing the stability
of the attractor solution in the decision module. We finally test these
results performing simulations of a spiking network where correlations among sensory neurons are now set by the micro-circuit balanced dynamics. Our results help to elucidate the contribution of feedforward and feedback mechanisms to
the generation of CP.
[1] Britten et al. J. Neurosci. 1992
[2] Nienborg and Cumming. Nature, 2009.
[3] Wang. Neuron, 2002. |
|---|
| responsibles | Levenes, Mongillo |
|---|
| |
|