UC-Irvine MURI Grant: Penn State Group Personnel: ◮ David Hunter, co-PI ◮ Ruth Hummel, graduate student in statistics department ◮ Duy Vu, graduate student in statistics department Broadly speaking, we’ve been working on computational methods for fitting ERGMs. Much of this work so far is in collaboration with Mark Handcock (co-PI) and/or in loose collaboration with Carter Butts (co-PI).
UC-Irvine MURI Grant: Penn State Group ◮ Ruth Hummel (David Hunter, advisor) is a statistics graduate student supported by the MURI grant. ◮ Her dissertation work involves algorithmic methods for improving approximate maximum likelihood estimation in ERG models for networks. 1 ◮ Basic goal: Maximize likelihood functions of θ of the form ( θ − θ 0 ) T g ( y obs ) − log E θ 0 exp { ( θ − θ 0 ) g ( Y ) } . ◮ Naive approximation by m ( θ − θ 0 ) T g ( y obs ) − log 1 � exp { ( θ − θ 0 ) g ( Y i ) } m i = 1 using a sample Y 1 , . . . , Y m does not work well. We are looking at new methods of improving this approximation. 1 A short paper she coauthored with Hunter and Handcock on this topic won a Dec. 2008 student paper competition.
UC-Irvine MURI Grant: Penn State Group ◮ Duy Vu (David Hunter, advisor) is a statistics graduate student who will be supported on the MURI grant as of summer 2009. ◮ He has not yet chosen a dissertation topic; so far, his work has been on MCMC algorithms for networks, which are vital in estimation and simulation of ERG models. ◮ Topic #1: Improving algorithms in the ergm package for sampling networks with (a) a fixed known degree distribution; (b) a fixed known degree for each node. ◮ Topic #2: Improving a simulated annealing-based sampling method in the ergm package to attempt to find a single network exhibiting a given set of network statistics.
Recommend
More recommend
