Title: Drifting Distributions

Abstract:  We will take a look at two space efficient, change detecting
algorithms: ADWIN and the sliding window.  These algorithms have various benefits
and shortcomings which will be discussed in some detail.  The objective of the
talk is to analyze these two change detectors with the goal of running them
alongside an online version of the Expectation-Maximization algorithm.  This will
serve the purpose of learning a drifting distribution in a very time and space
efficient manner.  We will briefly overview the EM algorithm and evaluate possible
online updates for the algorithm as well as a brief overview of ongoing and
future work for this learning method.