Convex Relaxations for Learning Bounded Treewidth Decomposable Graphs - Inria - Institut national de recherche en sciences et technologies du numérique
Conference Papers Year : 2013

Convex Relaxations for Learning Bounded Treewidth Decomposable Graphs

Abstract

We consider the problem of learning the structure of undirected graphical models with bounded treewidth, within the maximum likelihood framework. This is an NP-hard problem and most approaches consider local search techniques. In this paper, we pose it as a combinatorial optimization problem, which is then relaxed to a convex optimization problem that involves searching over the forest and hyperforest polytopes with special structures, independently. A supergradient method is used to solve the dual problem, with a run-time complexity of $O(k^3 n^{k+2} \log n)$ for each iteration, where $n$ is the number of variables and $k$ is a bound on the treewidth. We compare our approach to state-of-the-art methods on synthetic datasets and classical benchmarks, showing the gains of the novel convex approach.
Fichier principal
Vignette du fichier
convextjt_long.pdf (211.42 Ko) Télécharger le fichier
Origin Files produced by the author(s)
Loading...

Dates and versions

hal-00763921 , version 1 (11-12-2012)

Identifiers

Cite

K. S. Sesh Kumar, Francis Bach. Convex Relaxations for Learning Bounded Treewidth Decomposable Graphs. International Conference on Machine Learning, Jun 2013, Atlanta, United States. ⟨hal-00763921⟩
212 View
174 Download

Altmetric

Share

More