Max Planck Institute for Dynamics and Self-Organization -- Department for Nonlinear Dynamics and Network Dynamics Group
Personal tools
Log in


Tuesday, 12.06.2012 17 c.t.

Neuronal variability and its relation to perceptual decision making

by Dr. Jaime de la Rocha
from Institut D'Investigacions Biomèdiques August Pi i Sunyer, Barcelona, Spain

Contact person: Fred Wolf


Ludwig Prandtl lecture hall


Shared variability in the activity of cortical neurons has been proposed to impact perceptual decisions. This would explain the above-chance probability of predicting behavior from single neuron response fluctuations (choice probability, CP). Previous work on motion discrimination using Random Dot Patterns (RDP) has shown that trial-to-trial fluctuations in the realization of the RDPs can bias the choice with a time-course, quantified by the stimulus Motion Energy (ME), which rises after stimulus onset and decays back to zero after approximately 400 ms (Kiani et al. 2008). Paradoxically, fluctuations in the precise realization of RDPs were shown not to affect single neuron variability (Britten et al. 1993, Cohen & Newsome 2009). In this talk, I will address this paradox while seeking a unified understanding of how stimulus, neuronal, and behavioral fluctuations are related in plausible cortical circuits. I will show a new analysis of old data from a classical experiment (Britten et al. 1996). As previously shown, fluctuations in 0% coherence stimuli had no impact on the variability of the response of MT neurons defined over 2 s count windows, but we found that it had a substantial effect for smaller windows (e.g. 125 ms). Average Fano Factors were significantly larger for fluctuating (1.35, n=79) vs. identical stimuli (1.10, n=45). Average CP showed a fast increase followed by a plateau for fluctuating stimuli and exhibited a significant reduction during the first 400 ms for identical stimuli (mean CP across 0-500 ms was 0.53 and 0.50, respectively). To investigate which circuit mechanisms can simultaneously produce the time-courses of CP and ME profiles, I will present a network model of a sensory (MT) and a decision circuit (LIP). The attractor dynamics of LIP produce ME profiles with qualitatively the same behavior as the data. If identical stimuli are used, the average correlation between neurons can be very small - despite neurons in MT sharing a significant fraction of external and recurrent inputs (e.g. 20 %) - yielding only chance-level CPs. Promoting competition between sensory populations gave rise to a correlation structure which caused significant CP but its time-course mimicked the ME, i.e. with no sustained component. However, adding topographically organized top-down connections between LIP and MT can parsimoniously reproduce the experimentally observed time-course of CPs. Our findings suggest that CP can be largely explained by stimulus fluctuations and variability of top-down signals, questioning the interpretation that hard-wired correlations in sensory circuits having a causal impact on decisions.

back to overview