James M. Crawford
James Crawford received his Ph.D. in Computer Science from
The University of Texas at Austin
in 1989. His thesis was entitled
"Access-Limited Logic-- A Language for Knowledge Representation".
After spending a year at the University of Texas as a Post-Doctoral
fellow, he accepted a position at AT&T Bell Laboratories in
the Artificial Intelligence Principles Research Department, where he
spent two years. After that, he spent three years as a Research
Assistant Professor at CIRL,
and 5 years at i2 Technologies, in Irving, Tx. Dr. Crawford is
currently at Nasa Ames Research Laboratory in the Bay Area. His
research interests include satisfiability and scheduling problems,
knowledge representation, tractable inference, reasoning by symmetry,
and
reasoning about time and action.
An annotated list of papers is
available.
He co-chaired the satisfiability competition at the
"International Competition and Sympoisium on Satisfiability
Testing" held in Beijing China. The
guidelines
for the competition are available, as are the
benchmark problems.
Research
- Knowledge representation
- As part of my dissertation I developed a knowledge-representation
system called "Algernon". Algernon combines frame-based
knowledge-representation with inference rules and aims to provide
efficient inference while maintaining a clear semantics. Algernon has
been used to support Elaine Rich's natural language project at MCC, to
teach the expert system's class at the University of Texas, and to
support the building of a prototype system combining rule-based and
model-based reasoning at AT&T. Ongoing work is focusing on
re-implementing the rule engine (we are particularly interested in
investigating rule compilation), and on understanding the relationship
between the inferential limitations built into Algernon (for
computational reasons) and inferential limitations in human
reasoning.
- Satisfiability
- Determining whether a propositional theory is satisfiable is a
prototypical example of an NP-complete problem. Furthermore, many
problems that occur in knowledge representation, learning, planning,
and other areas of AI are essentially satisfiability problems. Much
interesting experimental work has recently been done in AI on
randomly-generated satisfiability problems. It appears that as
problems become more constrained (i.e., the clause/variable
ratio is increased) there is a "cross-over point" at which the
probability of a randomly-chosen theory being satisfiable drops quite
quickly from almost one to almost zero. Further, the problems around
this cross-over point seem to be quite hard. We have found empirically
that, for 3-SAT, the number of clauses at the crossover point is a
linear function of the number of variables. There is much
additional work to be done here , e.g., trying out new ideas for
speeding up satisfiability checking, translating from LISP to very
fast C code, applying these ideas and results to more practical
problems, etc.
The code for tableau is
available for research purposes as a tar archive. This
includes both the SAT checker tableau (more or less as described in
Experimental Results on the Crossover Point in Random 3SAT. but
without the data-structures tuned for random 3SAT) and a
polynomial-time preprocessor ("compact") that may be useful in other
systems. Warning: This is research code. Comments
are minimal and much of it was written before I even heard of ANSI
standard C.
Contact Information
James Crawford
Technical Area Lead
Autonomy & Robotics Area
NASA Ames Research Center
Phone: 650 604 1139
E-mail: james at crawford.net