Working from blueprints: evolutionary learning for design

When confronted by a problem, human designers often work forward from similar previously solved problems to solve the current problem. Based on this principle, we propose a new learning system especially suited for design using case-based reasoning principles to augment genetic algorithm search. When confronted with a problem we seed a genetic algorithm's initial population with solutions to similar, previously solved problems and the genetic algorithm then adapts its seeded population toward solving the current problem. Preliminary results on open-shop scheduling and rescheduling indicate the feasibility of this approach.