View article

The noisy-OR gate is an important tool for a compact elicitation of the conditional probabilities of a Bayesian network. An imprecise-probabilistic version of this model, where sets instead of single distributions are used to model uncertainty about the inhibition of the causal factors, is proposed. This transforms the original Bayesian network into a so-called credal network. Despite the higher computational complexity generally characterizing inference on credal networks, it is possible to prove that, exactly as for Bayesian networks, the local complexity to update probabilities on an imprecise noisy-OR gate takes only linear, instead of exponential, time in the number of causes. This result is also extended to fault tree analysis and allows for a fast fusion of the causal effects on models with an imprecise-probabilistic quantification of the initiating events.