One approach to develop software efficiently is to reuse existing software by modifying a part of it. However, modifying software will often introduce unexpected side effects into other parts of it. As a result, it costs much time and care to modify the software. So, in order to modify software efficiently, we have proposed a functional model to represent information about side effects caused by modification and a model based supporting system for modifying software. So far, however, an expert software developer must describe the entire functional model of the target software through the analysis of practical modifying processes. This will be an unnecessary burden on him. Moreover, the larger target software becomes, the harder the model construction becomes. Therefore, an automatic constructing method of the functional model is needed in order to solve this problem. So, this paper considers a method of acquiring useful interaction information by learning from training examples of modification. However, in our application domain, it seems that it is impossible to make complete domain theory and to prepare a large number or training examples in advance. Therefore, our learning method involves an integration of explanation-based learning (EBL) from positive examples of modification generated by the user and Similarity-based learning (SBL) from positive or negative examples generated by the user and the learning system. As a result, our method can acquire valid knowledge about the interaction from not so many examples under incomplete theory. Then, this paper presents a constructing method, in which our proposed learning method is incorporated, of a functional model. Finally, this paper demonstrates construction of the functional model in the domain of an event-driven queueing simulation program according to our learning method.

This paper concerns the issue of learning strategies for problem solvers from trace data. Many works on Explanation Based Learning have proposed methods for speeding up a given problem solver (or a Prolog program) by optimizing it on some subspace of problem instances with high probability of occurrences. However, in the current paper, we discuss the issue of identifying a target strategy exactly from trace data. Learning criterion used in this paper is the identification in the limit proposed by Gold. Further, we use the tree pattern language to represent preconditions of operators, and propose a class of strategies, called decision list strategies. One of the interesting features of our learning algorithm is the coupled use of state and operator sequence information of traces. Theoretically, we show that the proposed algorithm identifies some subclass of decision list strategies in the limit with the conjectures updated in polynomial time. Further, an experimental result on N-puzzle domain is presented.