A Decision-theoretic Approach to Adaptive Problem Solving

Jonathan S. Gratch, GERALD F. DEJONG, Nithin Mathews · 1994

This article argues that it is desirable, and possible, to construct general problem solving techniques that automatically adapt to the characteristics of a specific application. Adaptive problem solving is a means of reconciling two seemingly contradictory needs. On the one hand, general purpose techniques can ease much of the burden of developing a application and satisfy the oft argued need for declarative and modular knowledge representation. On the other hand, general purpose approaches are ill-suited to the specialized demands of individual applications. General approaches have proven successful, only after a tedious cycle of manual experimentation and modification. Adaptive techniques promise to reduce the burden of this modification process and, thereby, take a long step toward reconciling the conflicting needs of generality and efficiency. A principal impediment to adaptive techniques is the utility problem -- the realization that learning strategies can degrade performance un...

Read the paper · More papers on PaperTik