Hands on Tutorial: Bayesian Approaches to Movement Decoding and Behavior Konrad Körding1,2 and Ian Stevenson1 1Northwestern University, Department of Physical Medicine and Rehabilitation 2Northwestern University, Department of Physiology NCM 2009: April 30, 2009 3:00-4:30p (Waikoloa Beach Marriott Resort)

Material

• Full Tutorial (Slides, Handout, and Matlab Code)
• Contest

Abstract

   Over the last couple of years, Bayesian statistics have received significant attention in the field of motor control. Many applications of Bayesian statistics fall into one of two classes. (1) For the modeling of behavior Bayesian principles are used to predict how subjects combine pieces of information. (2) For the decoding of neural data Bayesian principles are used to combine information obtained from multiple neurons. An understanding of the underlying concepts may contribute to progress in motor control. We believe that the only way of really learning about a technique is by actually using it. Computational techniques have the advantage that they can readily be used on any personal computer. We have designed a hands-on tutorial that uses matlab where participants will both learn about neural decoding and about the Bayesian modeling of human behavior.

   The tutorial will start with Bayesian decoding of neuronal data. Participants will learn how to predict movement from spike data recorded from a behaving monkey. We will start plotting histograms for each of two movement directions. We will then convert these histograms into relevant probability distributions. We will then use Bayes rule to combine such probabilities from many neurons into a joint estimation. Such data analysis is possible for virtually any dataset obtained in electrophysiological laboratories. The tutorial will continue with Bayesian techniques for the modeling of behavior. The same techniques as in the first part of the tutorial will be used, but now probabilities do not come from noisy neurons but from noisy perception. We will multiply probability distributions, visually analyzing what cue-combination does to probability distributions. Participants will get to see that the same Bayesian principles used for neural data analysis can also account for some movement behavior. Altogether the tutorial will focus on operations that are visual and instructive, we will use no math that goes beyond simple fitting and Bayes rule. The focus is for participants to see that Bayesian techniques are useful for the analysis of neural data and of human behavior.

About the Authors