Publications
Research
Blog
Awards & Honors
Career
Miscellaneous
Contact Info
|
|
Keki Burjorjee
|
|
I'm a doctoral candidate in the
Computer Science Department at
Brandeis University.
The subject of my research is the genetic algorithm's remarkable, yet
mysterious capacity for adaptation. I've developed a new hypothesis about the
workings of this algorithm. My hypothesis departs from the
building block hypothesis at a fundamental level, and has significant
implications
for the fields of combinatorial optimization, machine learning, evolutionary
biology, and, of course, genetic algorithmics. This hypothesis is based on a
breakthrough in identifying the computational power
underlying the adaptive capacity of genetic algorithms.
Latest draft of my
dissertation
See Also:
Red
Dots, Blue Dots
|
|
|
Publications & Manuscripts |
| |
- On the Workings of
Genetic Algorithms: The Genoclique Fixing
Hypothesis [pdf]
Keki M. Burjorjee
Currently
unsubmitted
- Two Remarkable Computational Competencies of the Simple
Genetic Algorithm [pdf]
Keki M. Burjorjee
Currently unsubmitted
- The Fundamental Problem with the Building Block Hypothesis
[pdf]
Keki M. Burjorjee
Submitted
to the Evolutionary Computation Journal, April 7, 2009
- Towards a Sound Theory of Adaptation for
the Simple Genetic Algorithm, [pdf]
Keki M. Burjorjee
Technical Report, 2007
- Sufficient Conditions for
Coarse-Graining Evolutionary Dynamics,
[pdf], slides [pdf]
Keki M. Burjorjee
In Foundations of Genetic Algorithms
Conference IX, 2007
- A
General Coarse-Graining Framework for Studying Simultaneous
Inter-Population Constraints Induced by Evolutionary Operations,
[pdf],
extended version with proofs [pdf], slides [pdf]
Keki M. Burjorjee, Jordan B. Pollack
In
Proceedings of the 2006 Genetic and Evolutionary Computation
Conference
- Theme
Preservation and the Evolution of Representation,
[pdf]
Keki M. Burjorjee, Jordan B. Pollack
In Proceedings of
the 2nd Indian
International Conference on Artificial Intelligence, 2005
- Theme
Preservation and the Evolution of Representation,
[pdf]
Keki M. Burjorjee, Jordan B. Pollack
Proceedings of
the Theory of
Representation Workshop at the Genetics and Evolutionary Computation
Conference, 2005
|
|
Research |
Combinatorial optimization
problems abound in domains ranging from operations research to
electrical engineering to artificial intelligence. New
combinatorial optimization problems arise constantly. The
properties of most are poorly understood. Of the handful of
techniques that have been successfully used to tackle poorly
understood (i.e., black-box) combinatorial optimization problems,
genetic algorithms are perhaps the best known. Unfortunately,
despite the routine use of genetic algorithms for over three
decades, the adaptive capacity of these algorithms has not been
convincingly accounted for. The building block hypothesis—the
reigning explanation for this adaptive capacity—rests on
strong assumptions that have not been validated. Moreover,
there are several anomalies in the empirical literature that
cannot be explained by this hypothesis.
In my Ph.D. thesis I
present a new hypothesis—one that departs from the building block
hypothesis at a fundamental level. This new hypothesis is based on
the discovery that the simple genetic algorithm can perform a
remarkable form of sublinear computation which has a
straightforward connection to the general problem of interacting
attributes in data-mining. The increase in clarity conferred by
the new hypothesis promises to precipitate significant improvements in our
ability to use genetic algorithms to tackle hard combinatorial
optimization problems. By way of empirical support for my
hypothesis I describe what I consider to be the first of such
improvements—a tweak called clamping—and present the results of an
experiment in which the use of this simple tweak dramatically
improved the performance of a genetic algorithm on large, randomly
generated instances of the MAX 3-SAT problem
Visit my blog:
Hacking Evolution
|
|
Awards & Honors
|
| |
- Winner of the Tiny
GA competition (GECCO 2006)
- Awarded the Sproull Fellowship for "unusually strong
potential for graduate study" (University of Rochester, 2000)
- Awarded the Mary Evelyn Wells and Gertrude Smith
Prize for excellence in the study of undergraduate Mathematics (Vassar
College, 1998)
- Graduated with
departmental honors in mathematics, and computer science, and
general honors overall (Vassar College, 1998)
- Part of a team of 3 that placed 2nd in an all-India
high school programming competition (Computer Society of India, 1992)
|
|
Career |
My
Resume
|
|
|
|
Miscellaneous |
SpeedyGA:
A fast Simple Genetic Algorithm in Matlab.
|
|
|
|
Contact
Information |
phone: (617) 645-6582
email:
yyy@cs.brandeis.edu
(replace yyy with "kekib")
|
|
|
