This work is supported by award HR0011-05-1-0040 from DARPA/IPTO.
Research ObjectivesThe brittleness of first-order symbolic logic has lead to the ascension of statistics as the dominant paradigm in AI. But the statistical paradigm has been unable to support the componential knowledge and the rich combinatorial inference of symbolic logic. Successful cognitive computing will need to combine the power of logic with the robustness that underlies statistics. The research in this project explores approaches to knowledge representation and automated reasoning that employ the inferential power of first-order logic but are supported by a new semantics that embraces the real-world robustness of statistical inference. The approach that this project develops follows a new form of Explanation-Based Learning, and acquires both EBL concepts and Partitioned Axiom Sets. If successful, we believe the research will lead to new algorithms that efficiently appreciate and adapt to their deployment contexts. They will learn quickly to supply a new user what is most needed to optimize their joint problem-solving behavior.
Eyal Amir Jerry DeJong Last updated on January 13, 2005.