Conditioning Methods for Exact and Approximate Inference in Causal Networks

We present two algorithms for exact and ap­ proximate inference in causal networks. The first algorithm, dynamic conditioning, is a re­ finement of cutset conditioning that has lin­ ear complexity on some networks for which cutset conditioning is exponential. The sec­ ond algorithm, B-conditioning, is an algo­ rithm for approximate inference that allows one to trade-off the quality of approxima­ tions with the computation time. We also present some experimental results illustrating the properties of the proposed algorithms.