The basic assumption underlying the research field of case-based reasoning (CBR) is that similar problems do have similar solutions. When being faced with a new problem, the core idea of CBR is to turn to a collection of past experiences, so-called cases, search for one that appears to be similar to the current one and to re-use (possibly with slight adaptations) the solution part of the former case. Thus, CBR represents an approach to building knowledge-based systems that, in part, mimics the way humans do inference when solving new problems.
Case-based reasoning is both, a powerful method for reasoning with computers, as well as a description of human behavior in everyday problem solving. With respect to the former, CBR has been formalized as a four-step process consisting of the stages retrieve, revise, reuse, and retain. A crucial challenge in case-based reasoning is the modeling and usage of appropriate similarity measures which are an essential part of the first phase. Using machine learning techniques for learning and optimizing such measures is a promising approach combining the strengths of data-intensive supervised machine learning and the intuitiveness, human-centered and explanative power of case-based inference.