Modeling causal reinforcement and undermining with noisy-AND trees

Abstract

When data are insufficient to support learning, causal modeling, such as noisy-OR, aids elicitation by reducing probability parameters to be acquired in constructing a Bayesian network. Multiple causes can reinforce each other in producing the effect or can undermine the impact of each other. Most existing causal models do not consider their interactions from the perspective of reinforcement or undermining. We show that none of them can represent both interactions. We present the first explicit causal model that can encode both reinforcement and undermining and we show how to use such a model to support efficient probability elicitation. © Springer-Verlag Berlin Heidelberg 2006.