Reproduce “Learning to Bound: Monotonic, Calibrated Neural Upper Bounds for MPE Inference” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Akshay Vyas, Hailiang Dong, Shivvrat Arya, Brij Malhotra, Tahrima Rahman, Vibhav Gogate, Nicholas Ruozzi
L2B tackles repeated MPE search in probabilistic graphical models by learning an upper-bound predictor specialized to each fixed model. It trains on costly offline labels, constrains predictions to decrease as evidence is added, and calibrates them for high-probability validity. On g80.f10, SMNN improved the MBE-i4 score by 1.34% at ten minutes; calibration is per node, not a whole-path guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.