Belief Change in OCF-Based Networks in Presence of Sequences of Observations and Interventions: Application to Alert Correlation

Abstract

Ordinal conditional function (OCF) frameworks have been successfully used for modeling belief revision when agents’ beliefs are represented in the propositional logic framework. This paper addresses the problem of belief change of graphical representations of uncertain information, called OCF-based networks. In particular, it addresses how to revise OCF-based networks in presence of sequences of observations and interventions. This paper contains three contributions: Firstly, we show that the well-known mutilation and augmentation methods for handling interventions proposed in the framework of probabilistic causal graphs have natural counterparts in OCF causal networks. Secondly, we provide an OCF-based counterpart of an efficient method for handling sequences of interventions and observations by directly performing equivalent transformations on the initial OCF graph. Finally, we highlight the use of OCF-based causal networks on the alert correlation problem in intrusion detection.