The progress in understanding of the theory of default logic has been reflected in the development of algorithms for computing default extensions, as well as in first implementation projects.
Thus, we contend that default logic can not only serve as a declarative knowledge representation formalism but has also the potential to evolve into a practical high-level computational environment. This can be accomplished through the development of much faster than presently available implementations of default logic, through the systematic studies and development of the methodology of programming with default logic, through comprehensive experimentation and by demonstrating usefulness of default logic as a computational mechanism for AI applications. These are precisely the main objectives of our research proposal.
More specifically, to improve performance of default reasoning, we investigate search optimization techniques for backtracking that are used in solving main default reasoning tasks. In addition, theoretical results on algorithms and complexity of default reasoning indicate that parallel and distributed implementations of default reasoning may greatly enhance the performance. We are following this direction.
We also study algorithms for restricted fragments of default logic, especially, for logic programming with stable model semantics. This direction is quite promising as Davis-Putnam-like methods for computing stble models were recently discovered.
Next, although default reasoning has a distinctly declarative character, no systematic methodology for programming and problem solving with default logic is known. It is necessary to establish such methodology, or, in effect, guidelines for a default logic programmer. We are investigating such principles.
In order to experiment with default reasoning one needs access to a large number of automatically generated classes of default theories and logic programs. We are developing a system which will satisfy these requirements. It will facilitate experimental work and benefit entire automated reasoning community.
Our research will transform default logic into an effective computational tool. It will:
We expect that due to the declarative nature of default logic, this new programming tool will be easy to use, will allow for fast prototyping and will be effective in a wide range of artificial intelligence applications.
Funding for this project has been requested from the NSF.