Date: Feb 20th, Tuesday
Time: 3:00 pm
Place: Little 339 (the Atrium)
Speaker: Pengwen Chen

Title: Make metrics from Bregman divergences
　

Abstract: Bregman divergences become one important tool in the
 research area of machine learning, e.g. in clustering, vector
 quantization.Most Bregman divergences are not metrics.
Metric property provides a valuable information in pattern recognition research
area. It is known that the square root of Jensen-Shannon divergence is a metric.
In fact, Jensen-Shannon divergence can be regarded as a special case of
averaging Bregman divergences with an associated convex function
$x\log x$. It is very natural to ask whether their square root is
also a metric for other Bregman divergences?

This is the main motivation of this talk. I will provide a sufficient and
necessary condition on the associated convex functions, such that the square
root of averaging Bregman divergence is a metric.
　

The pizza and drinks will be provided after the talk.