Detailed Information About Manuscript entitled, "Correlated
Equilibria and Probabilistic Inference in Graphical Models"
- A conference version of Section 3, entitled "Graphical
Potential Games," was first submitted to WINE 2010
in August 2010. It was latter submitted to NIPS 2014 and AAAI
2015. In April 2015, I also submitted that work to The 26th International Conference on
Game Theory (ICGT15). After acceptance for
presentation at ICGT15, I created an arXiv version of the
submission because ICGT15 made the paper public via
web posting. Then I submitted the paper one last time to NIPS 2015. I presented
the work at ICGT15 in July 2015. I submitted
a journal
version of the manuscript to GEB on September 24, 2015 (rejected:
March 18, 2016).
- Section 4 contains the theoretical work
behind the application of algorithms for computing equilibria in
games to the problem of belief inference in probabilistic graphical
models. Ze Gong, an MS student working on his MS project under my
supervision carried out preliminary
experiments to demonstrate the feasibility of the approach using
simple versions of no-regret algorithms from the literature on
learning in games. In an upcoming manuscript, entitled "Correlated Equilibria
for Approximate Variational Inference in MRFs," I extend this work. The
extended version now also includes (1) an additional more global
approach
based on using a version of fictitious play on a two-player
potential game; and (2) a more exhaustive experimental section illustrating the
effectiveness of the proposed techniques in higher-dimensional Ising
models with grid graphs. We compare the proposed algorithms with the
most popular, commonly used, and state-of-the-art techniques with
simple implementations. I expect to submit a shorter conference version of
that manuscript for review to NIPS 2016.
- Section 5 formally explains how the linear-programming
formulations in my paper with Sham Kakade,
Michael Kearns, and John Langford, "Correlated Equilibria in
Graphical Games" (EC'03), extends easily to
bounded-tree-width and polymatrix graphical games, using the same original
approach. I have
included related comments in two later manuscripts. One is an arXiv technical report entitled, "On Sparse Discretization
for Graphical Games." The other is an arXiv technical
report entitled, "FPTAS for
Mixed-Strategy Nash Equilibria in Tree Graphical Games and Their
Generalizations," with Mohammad Irfan.
Last modified: Fri Apr 8 22:46:57 EDT 2016