Barrier Frank-Wolfe for Marginal Inference

Abstract

We introduce a globally-convergent algorithm for optimizing the tree-reweighted (TRW) variational objective over the marginal polytope. The algorithm is based on the conditional gradient method (Frank-Wolfe) and moves pseudomarginals within the marginal polytope through repeated maximum a posteriori (MAP) calls. This modular structure enables us to leverage black-box MAP solvers (both exact and approximate) for variational inference, and obtains more accurate results than tree-reweighted algorithms that optimize over the local consistency relaxation. Theoretically, we bound the sub-optimality for the proposed algorithm despite the TRW objective having unbounded gradients at the boundary of the marginal polytope. Empirically, we demonstrate the increased quality of results found by tightening the relaxation over the marginal polytope as well as the spanning tree polytope on synthetic and real-world instances.

25 pages, 12 figures, To appear in Neural Information Processing Systems (NIPS) 2015, Corrected reference and cleaned up bibliography

Cited by in corpus (9)

• On the Global Linear Convergence of Frank-Wolfe Optimization Variants
• LP-SparseMAP: Differentiable Relaxed Optimization for Sparse Structured Prediction
• Learning with Fenchel-Young Losses
• Safe Screening for the Generalized Conditional Gradient Method
• Arbitrage-Free Combinatorial Market Making via Integer Programming
• Learning Infinite RBMs with Frank-Wolfe
• Approximate Frank-Wolfe Algorithms over Graph-structured Support Sets
• Convex Optimization For Non-Convex Problems via Column Generation
• Lost Relatives of the Gumbel Trick