Context-sensitive Network: A Probabilistic Context Language for Adaptive Reasoning

Decision support systems based on probabilistic and decision-theoretic foundations are becoming a popular choice in a number of domains, ranging from medicine to biology, human-computer interaction and robotics. Representing and reasoning with scalable and adaptable context-sensitive information in decision support systems is becoming an important problem because the relationships or the associations of an object with other objects are often context-based. For instance, a gene may influence or regulate another gene only if a particular molecule or context is in an activated state. The problem is difficult because we may not know the number of objects (genes), contexts (molecules), and their relevant relationships at the design time. Furthermore, the increase in the number of uncertain context attributes may lead to highly complex networks where the inference may be intractable.

We introduce a new probabilistic graphical language, Context-Sensitive Network (CSN), which extends Bayesian Networks by incorporating context value-dependent asymmetry for compact and accurate representation of uncertain knowledge. We describe the theoretical foundations and the practical considerations as well as an empirical evaluation of the representation, inference, and learning in the proposed language. Our representation is particularly useful when there are a large number of relevant context attributes, when the context attributes may vary in different conditions, and when all the context values or evidence may not be known beforehand.