Computing Upper and Lower Bounds on Likelihoods in Intractable Networks

This addresses the challenge of probabilistic inference in large or complex networks where exact computation is infeasible, though it appears incremental as it builds on existing bounding methods.

The paper tackles the problem of computing marginal probabilities in intractable sigmoid and noisy-OR networks by developing deterministic techniques for upper and lower bounds, with results demonstrated through numerical experiments showing tightness.