WHO:	Peggy Series , Postdoctoral Fellow working in Pouget Lab
TOPIC:	Orientation Selectivity, Tuning Curve Sharpening, and the Efficiency of a Population Code
ABSTRACT:	Many neurons in cortex exhibit bell-shaped tuning curves. Several studies have shown that the information conveyed by these tuning curves increases as their width decreases, leading to the notion that sharpening tuning curves improves population codes-- a principle invoked by models in a wide variety of domains, including orientation selectivity, perceptual learning, attention, and auditory processing. This notion, however, is based on the assumptions that the noise distribution (trial-to-trial variability) is 1) independent among neurons, and 2) independent of the tuning curve width. We have reexamined these assumptions in networks of spiking neurons using orientation selectivity as an example. We compared the two major classes of models: a model in which the tuning curves are sharpened through cortical lateral interactions and a model in which the orientation selectivity is the result of the convergence of LGN afferences with no further sharpening in the cortex. We report that the sharpening model conveys far less information about orientation than the no-sharpening model. Therefore, contrary to what is widely assumed, sharpening through lateral connections in networks of spiking neurons leads to severe losses of information. Moreover, the code produced by the sharpening model is particularly inefficient for learning and computation because the majority of the information is conveyed by correlations. This work also makes several experimental predictions that could be used to distinguish between the sharpening and no-sharpening models.
WHEN:	3/5/2004 12:00:00 PM
WHERE:	Meliora 269

Events Homepage