Nonsystematic Methods

Suppose you have left your keys somewhere in the house but cannot remember where, and need to find them quickly. You could do a systematic search of the entire house, perhaps starting in the garage and exploring every nook and cranny to be absolutely sure the keys are not there before moving to other rooms. However, experience suggests that this ought to be a last resort and that a quick search following your instinct or ``following your nose'' is often likely to be successful more quickly.

This suggests that when the search space is very large and a complete search will take too long, then it can be worthwhile to give up any hope of ever doing a complete systematic search and instead use an algorithm that concentrates on first exploring a good selection of the most likely places.

Many such algorithms are based on some form of ``iterative repair.'' We propose a solution, look at its deficiencies and then try to repair one of these deficiencies, obtaining a new state that we hope is closer to an acceptable solution. This process is repeated many times until the state is acceptable, or we run out of patience. In order to save memory such algorithms cannot remember all the places that have already been searched. Hence, they often intentionally involve some degree of randomness: if decisions are made by tossing a coin then it is less likely that we will repeat exactly the same mistakes on each attempt to solve the problem.

CIRL

CIRL's work on larger practical problems has lead to the development of powerful nonsystematic methods such as Squeaky Wheel Optimization and Schedule Packing. Our work on WSAT has focussed on understanding its performance, and extending it to handle larger problems.

Pointers

Parent areas:
Search

Subareas:
WSAT
Squeaky Wheel
Schedule Packing

Back to the CIRL Overview Page.