Exploiting Causal Independence in Bayesian Network Inference