|
|
Publications
Blog
Miscellaneous
Career
Contact Info
|
Keki Burjorjee |
|
|
I'm a Ph.D. student at the DEMO Lab in the Computer Science Department at Brandeis University. Natural evolution can
be thought of as a computational process. I'm interested in
identifying the computational principles underlying the remarkable adaptive
power of this process, and in using this form of computation to solve challenging optimization
and machine learning problems. |
|
|
Publications |
|
|
|
|
|
I am fascinated by the way natural systems solve hard computational problems. During my graduate work, first at the Brain and Cognitive Sciences Department at the University of Rochester, and more recently at the DEMO Lab at the Computer Science Department at Brandeis University, I have learned techniques for investigating the unconventional ways in which natural systems compute. Natural systems such as brains and evolutionary systems have abilities that far outstrip those of man-made systems in several important domains. A better understanding of the unconventional ways in which such systems compute is likely to permit AI engineers to transcend some of the fundamental limitations they experience when attempting to use conventional computing techniques to solve hard problems. My research in the past few years has centered on the genetic algorithm. I have developed a new mathematical framework for analyzing the short-term dynamics of these algorithms, and a new technique (in Matlab) for visualizing their high-dimensional dynamics. Using these tools I have identified a powerful type of computation that genetic algorithms can perform efficiently, scaleably and robustly. This discovery suggests a new explanation for the adaptive ability of genetic algorithms which is radically different from the dominant hypothesis of the field (the building block hypothesis). This new explanation undermines the fundamental assumptions under which genetic algorithms, and indeed other types of recombinative evolutionary algorithms, are typically researched and applied; it also suggests new ways to improve the adaptive ability of genetic algorithms. |
|
|
Some Recent Blog Posts |
|
VectorGA: An implementation of a genetic algorithm in Matlab. |
|
My Resume |
|
| yyyy@cs.brandeis.edu (replace yyyy with kekib) |
|
|
|
|
| Last Updated 2008-06-17 |
|