Case-based reasoning for autonomous constraint solving

Humans often reason from experiences in the way exemplified above. Faced with a new problem, we recall our experiences in solving similar problems in the past, and we modify the past solutions to fit the circumstances of the new problem. Within Artificial Intelligence (AI), the idea that we can solve problems by recalling and reusing the solutions to similar past problems, rather than reasoning 'from scratch', underlies Case-Based Reasoning (CBR), which has been the target of active research and development since the late 1980s. CBR is a problem solving and learning strategy: reasoning is remembered (this is learning); and reasoning is remembering (this is problem-solving). CBR can be useful in domains where problem types recur, and where similar problems have similar solutions. Its wide range of application areas - from classification and numeric prediction to configuration, design and planning - and domains - from medicine to law to recommender systems - is testimony to its generality. In this chapter, we review the application of CBR to search and especially to constraint solving. We present CPHYDRA, a recent successful application of CBR to autonomous constraint solving. In CPHYDRA, CBR is used to inform a portfolio approach to constraint problem solving.