Time: October 8, 2009, 1125-1150 hours
Place: Erskine 415
Speaker: Jennifer Harlow

Title:
Markov Chains over Regular Subpaving Histograms

Abstract:
We extend a class of binary tree partitions of a hyper-cube called 
regular subpavings into a class of simple functions.  Addition and scalar 
multiplication are extended to this class of simple functions in order 
to efficiently compute the sample mean of a Markov chain whose state space 
is this class.  The Markov chain is used to obtain the posterior mean over 
a space of adaptive histograms in the context of density estimation for 
massive data problems under memory constraints.