Join-graph based cost-shifting schemes

TL;DR

This paper introduces hybrid algorithms combining minibucket elimination and message-passing for bounding MPE queries in graphical models, demonstrating their effectiveness through empirical evaluation and winning a challenge.

Contribution

It presents novel hybrid algorithms that leverage two common bounding techniques, offering improved performance and practical success in inference tasks.

Findings

Hybrid algorithms outperform individual methods in experiments.

A heuristic from our algorithms led to winning a PASCAL2 inference challenge.

The approach balances efficiency and accuracy in graphical model inference.

Abstract

We develop several algorithms taking advantage of two common approaches for bounding MPE queries in graphical models: minibucket elimination and message-passing updates for linear programming relaxations. Both methods are quite similar, and offer useful perspectives for the other; our hybrid approaches attempt to balance the advantages of each. We demonstrate the power of our hybrid algorithms through extensive empirical evaluation. Most notably, a Branch and Bound search guided by the heuristic function calculated by one of our new algorithms has recently won first place in the PASCAL2 inference challenge.