RI: Small: Scaling Up Inference in Dynamic Systems with Logical Structure

University Of Illinois At Urbana-Champaign, Urbana IL

Investigators

Abstract

Stochastic inference problems arise naturally in many applications that interact with the world, such as natural language processing (NLP), robot control, analysis of social networks, and environmental engineering. Current real-world applications require inference mechanisms that can scale to thousands and more interactions. This project is scaling up and improving precision of automated inference and learning in dynamic partially observable domains, with later application to decision making, by combining the complementary computational strengths of logical and probabilistic methods. The inference and learning algorithms being developed are being assessed by theoretical and experimental means, with aims of improving question answering from text narratives and environment-state estimation for mobile robots. Examples of societal benefits are enabling people to access and examine details hidden in large amounts of textual information as well as enabling more helpful and autonomous mobile robots.