Testing efficient coding for a complete and inhomogeneous neural population
Wednesday 24th of November 2010 at 12:00pm
The theory of efficient coding posits that the bandwidth of communication channel in the neural system should be maximally utilized for information transmission. This idea has been a guiding principle for understanding the structure and function of sensory systems. It is difficult, however, to test the theory. In the earlier studies, the validity has relied on the visual impression of similarity between the theoretical predictions and the data. Here we present extensions of the earlier theoretical models that enable a direct and quantitative test of efficient coding with physiology data recorded from an entire neural population in a small retinal patch. We conclude that the retinal linear transform is about 80% optimal for information transmission. We also find that the optimal and retinal linear transforms yield highly redundant population codes of natural images, suggesting that the redundancy reduction principle, a special form of efficient coding, is too simplistic to understand the retinal processing. Our extended model is general and can be applied to any neural systems, providing a useful tool to assess the optimality of a linear transform with an entire neural population.
508-20 Evans Hall
This is a joint work with Jeff Gauthier, Greg Field, Alexander Sher, John Shlens, Martin Greschner, Tim Machado, Keith Mathieson, Deborah Gunning, Alan Litke, Liam Paninski, EJ Chichilnisky, and Eero Simoncelli.
Join Email List
You can subscribe to our weekly seminar email list by sending an email to
firstname.lastname@example.org that contains the words
subscribe redwood in the body of the message.
(Note: The subject line can be arbitrary and will be ignored)