Bayesian Inference with Probabilistic Population Codes
Speaker: Dr. Alexandre Pouget , University of Rochester
Relevant URL: http://www.csail.mit.edu/events/eventcalendar/calendar.php?show
Recent psychophysical experiments indicate that humans perform near-optimal Bayesian inference in a wide variety of tasks, ranging from cue integration to decision-making to motor control. This implies that neurons both represent probability distributions and combine those distributions according to a close approximation to Bayes rule. At first sight, it would appear that the high variability in the responses of cortical neurons would make it difficult to implement such optimal statistical inference in cortical circuits. I will show that, in fact, this variability generates probabilistic population codes which represent probability distributions over the encoded stimulus. Moreover, when the neural variability is Poisson-like, as is the case in cortex, a broad class of Bayesian inference, such as cue integration, or integrating evidence for decision-making, can be closely approximated with simple linear combinations of probabilistic population codes. Therefore, this theory suggests that the Poisson-like variability in the cortex greatly simplifies Bayesian inference in neural circuits.