Eizaburo Doi

Characterization of optimal population coding with the linear Gaussian model

Tuesday 23rd of November 2010 at 01:00pm
508-20 Evans Hall

Eizaburo will be leading an informal discussion of the following topic.

Efficient coding is commonly defined by maximizing information transmission subject to the channel’s bandwidth. Efficiency can be more generally defined with other objective functions and constraints. In this talk, we will characterize the solutions with various objectives and constraints under the linear Gaussian model. More specifically, we examine the solutions with two objectives: 1) maximizing mutual information, 2) minimizing mean squared errors; subject to three kind of constraints: a) number of coding units (neural population size), b) bandwidth (filter’s output power), c) weights (filter’s L2 norm). In the linear Gaussian model, there are two kind of noise (input and output noise), and we clarify how their amplitudes shape the solutions in different settings.

This is a joint work with Liam Paninski, Doru Balcan, Mike Lewicki, and Eero Simoncelli.

Join Email List

You can subscribe to our weekly seminar email list by sending an email to majordomo@lists.berkeley.edu that contains the words subscribe redwood in the body of the message.
(Note: The subject line can be arbitrary and will be ignored)