Efficient Pathfinding in Very Large Data Spaces
Abstract
This project created a corpus of large test problems relevant to the Intelligence Community (IC). These problems required bringing together only facts and rules from the unclassified OpenCyc knowledge base. Because many current theorem provers are unable to a reason with (or even load) an IC-sized knowledge base, six different levels of problems were created, each containing progressively larger theories. Dozens of Automatic Theorem Proving (ATP) researchers are now heavily engaged in attacking more and more of these six new TPTP problem sets. In addition to challenging the theorem-proving community, this project contained a series of experiments to assess and, where possible, improve the efficiency of Cyc s general inference engine. These experiments identified areas for immediate improvement, and approximately one full factor of 10 speedup was obtained just in the course of carrying them out and analyzing their results.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 2007
- Accession Number
- ADA475387
Entities
People
- Douglas B. Lenat
- Keith Goolsbey
- Kevin Knight
- Pace Smith