BSF:2012144:Human-Competitive Evolutionary Computation

Hampshire College, Amherst MA

Investigators

Abstract

This project is funded as part of the United States-Israel Collaboration in Computer Science (USICCS) program. Through this program, NSF and the United States - Israel Binational Science Foundation (BSF) jointly support collaborations among US-based researchers and Israel-based researchers. This project will examine and extend computational techniques that allow computers to out-perform humans on certain classes of difficult problems, some of which have practical applications across many areas of science and engineering. The techniques to be studied, which are modeled on biological processes of evolution, have already been shown to produce human-competitive performance in areas ranging from pure mathematics to quantum system design, and from game-playing systems to software engineering and debugging. In this project the co-PIs will collaborate to characterize the application areas in which these human-competitive successes have been achieved, along with the specific techniques that were used in each case. The project will use these characterizations to guide improvements to the techniques and applications to new problems. The initial data for the study will be taken from the winners of the of the Human Competitive Results Competition that has been held annually since 2004 at the Genetic and Evolutionary Computation Conference. Areas of anticipated application work include software engineering, for example for automatic program synthesis and repair, where early successes and the scale of the potential applications indicate substantial promise. The long-term potential of work in this area is for computers to automatically generate useful software, for many applications of scientific or social significance, that would be prohibitively difficult or expensive for human programmers to produce. The project will be conducted with undergraduates and graduate students and will also thereby help to train future innovators in computer science.