A Hybrid Approach for Probabilistic Inference using Random Projections

We introduce a new meta-algorithm for probabilistic inference in graphical models based on random projections. The key idea is to use approximate inference algorithms for an (exponentially) large numb