Optimal encoding and decoding in sensory populations
Add to your list(s)
Download to your calendar using vCal
If you have a question about this talk, please contact Dr Máté Lengyel.
Experimental evidence suggests that human judgments of many perceptual attributes are consistent with Bayesian estimation, in which noisy sensory measurements are combined with prior knowledge of the environmental distribution of those attributes. How are such computations achieved in the brain? I’ll first describe a means by which populations of neurons can efficiently encode a scalar sensory variable. The structure of this representation, which can be derived in closed form based on the prior distribution of the relevant sensory variable, provides an implicit encoding of the prior, and is consistent with both physiological and perceptual data for a variety of sensory attributes. I’ll then describe a novel decoder that can approximate the Bayes least squares estimate, converging to the true value as the neural population size increases. The decoder is neurally plausible, and requires knowledge only of the preferred stimuli and a fixed filter, and not the prior distribution or family of tuning curves. I’ll end by discussing the means by which the neural populations might go about learning these representations.
This talk is part of the Computational and Biological Learning Seminar Series series.
This talk is included in these lists:
Note that ex-directory lists are not shown.
|