Reasoning

Complementing general questions of how to represent knowledge is the need to understand how knowledge can be used. In general, realistic problems have enormous associated spaces of possible solutions which must be explored (searched) to find an actual solution that meets the requirements of the problem. These spaces are much too large to be searched in their entirety, and ways must be found to focus or short-circuit the search for solutions if systems are to have any practical utility.

Put another way, almost every interesting problem class is computationally intractable (NP-complete or worse), meaning that in the worst case problems may take time exponential in the problem size to solve.

Important questions include ways to reduce problem size and to focus search, and techniques for finding approximate solutions quickly.

CIRL

CIRL is working on reasoning on many fronts, including systematic and nonsystematic search and polytime limited reasoning methods. An important focus of our work is how KR systems can be made to be usable for solving real problems, for example by exploiting structure in the problems themselves or in the associated solution space.

Pointers

Subareas:
Search
Tractable Reasoning
Back to the CIRL Overview Page.