Advanced Approximation Methods and Specification Schemes for Automated Reasoning

University Of California-Irvine, Irvine CA

Investigators

Abstract

This project will explore three avenues in reasoning and knowledge representation: development of new approximation methods that incorporate user-adaptive and any-time features; development of hybrid knowledge-bases that combine deterministic information (constraints) and probabilistic information (belief networks), and which are both semantically coherent and computationally effective; and application of hybrid languages and algorithms to temporal reasoning problems in the domains of planning, scheduling, and diagnosis. The outcome of this research will include a system of algorithmic tools which address issues of non-tractability in an innovative and practical manner, and which are applicable to a new knowledge-based framework that allows the expression of both causal and constraint-like information, thus facilitating tasks such as planning, diagnosis and design. Parameterization will allow users to control the algorithms and adjust them to their own domains and resources. The computational tools will support the solution of challenging problems at the frontiers of diverse areas of science and industry such as robotics, planning and scheduling, bioinformatics (linkage analysis and protein secondary structure prediction), and e-commerce (multi-agent combinatorial auctions).