Decomposition bounds for marginal MAP

摘要: Marginal MAP inference involves making predictions in systems defined with latent variables or missing information. It is significantly more difficult than pure marginalization and tasks, for which a large class of efficient convergent variational algorithms, such as dual decomposition, exist. In this work, we generalize decomposition to generic power sum task, includes marginal MAP, along special cases. Our method based on block coordinate descent algorithm new convex bound, that guaranteed converge monotonically, can be parallelized efficiently. We demonstrate our approach queries real-world problems from the UAI approximate challenge, showing framework faster reliable previous methods.