Reducing the Adaptation Overhead

Déjà Vu is a case-based software design system whose success is based on two novel techniques: hierarchical case-based reasoning and adaptation-guided retrieval. Hierarchical case-based reasoning (HCBR) solves complex problems in a hierarchical fashion by reusing and combining the solutions of many separate cases and is fully supported by the normal CBR retrieve-and-adapt cycle. Adaptation-guided retrieval (AGR) makes use of adaptation knowledge during retrieval to find the most adaptable case to the target in the case-base and to make predictions about the type of adaptations that will be needed to reuse that case. This technique overcomes the failures that can occur in traditional retrieval methods using semantic-similarity. The paper will describe HCBR and AGR in the context of Déjà Vu's plant-control software design domain. Experimental evidence is provided to support our claims of improved problem solving performance and flexibility.