<html>
<body>
<div align="center"><font size=4><b>The UCSD Department of Cognitive
Science is pleased to announce a talk by<br><br>
</font><font size=5>Wei Ji Ma Ph.D.<br><br>
</font>Department of Brain and Cognitive Sciences<br>
University of Rochester<br><br>
<font face="arial">Tuesday, January 22, 2008 at 2pm<br>
Cognitive Science Building, room 003<br><br>
<br>
</font><font size=5>"The neural basis of probabilistic computation
in perception."<br><br>
</b></font></div>
Perception often requires the combination of multiple pieces of uncertain
information across sensory modalities (e.g. speech recognition), time
(e.g. decision-making), or space (e.g. visual search, change detection).
For many such tasks, human behavior has been shown to approach
statistical optimality. This raises the question how optimal computation
is implemented in biologically realistic neural networks. Using the
framework of population coding, I will show that optimal cue integration
– a common aspect of multisensory perception – is realized through linear
operations on patterns of activity, provided that neural variability
belongs to a broad family of distributions which we call Poisson-like.
This suggests that the form of neural variability plays a crucial role in
facilitating optimal computation. I will discuss predictions for
physiological experiments. I will then outline a research program which
aims to understand human behavior in and neural implementation of a wide
variety of probabilistic computations, as well as deviations from
optimality.<br>
</body>
</html>