Liam Paninski, 10/7/03
<br><br>

Inferring beliefs from Bayes-optimal behavior
<br><br>

I'll discuss a method for inferring a subject's beliefs (in the form
of an a priori distribution) from his/her behavior, under the
assumption that the behavior is (near-) optimal from a Bayesian
perspective.  The problem can be cast in a fairly nonparametric way
(i.e., we don't have to impose any strong assumptions on the subject's
beliefs), and the solution is surprisingly computationally tractable.
<br><br>

Time permitting, I'll also discuss some recent results on
maximum-likelihood estimation of a point process neural model, for
which some of the mathematics turns out to have a similar flavor.