Model Failure and Context Switching Using Logic-Based Stochastic Models

Abstract We define a notion of context that represents invariant, stable-over-time behavior in an environment and we propose an algorithm for detecting context changes in a stream of data. A context change is captured through model failure when a probabilistic model, representing current behavior, is no longer able to fit the newly encountered data. We specify stochastic models using a logic-based probabilistic modeling language and use its learning mechanisms to identify context changes. We also discuss how our algorithm can be incorporated into a failure-driven context-switching probabilistic modeling framework and demonstrate several examples of its application. Keywords Probabilistic reasoning

• Type: Journal
• Year: 2010
• Where: JCST
• Authors: Nikita A. Sakhanenko, George F. Luger
• Keywords: Context Changes, JCST 2010, Probabilistic, Probabilistic Modeling