Unlock how to turn complex, non-deterministic problems into guided solutions.
This non-fiction work explains a strategy to reduce the heavy costs of non-deterministic programming by learning control heuristics from traces of sample solutions.
It introduces CRAPS, a language for describing patterns found in sequences of knowledge applications. By analyzing traces generated with human guidance, the authors show how to convert these patterns into a control framework that helps solve future problems more efficiently. The book discusses the utility of this approach, offers a concrete example, and considers the limits of the technique when training data is imperfect.
Readers will see how a production-style program can be guided from many possible paths to a more deterministic workflow. The discussion covers related ideas like meta-rules and probabilistic guidance, and it explores how these concepts might evolve to improve declarative knowledge bases and problem-solving systems.
- How declarative knowledge can be coupled with simple control patterns to reduce search costs
- The creation of CRAPS descriptions from traces of rule applications
- A practical example using a jigsaw-puzzle-style problem to illustrate learning heuristics
- Considerations about training quality, data dependence, and future directions for learning control information
Ideal for readers of artificial intelligence, machine learning, and knowledge representation who want a concrete approach to learning and applying problem-solving heuristics.