Domain-based program synthesis using planning and derivational analogy

In this thesis, I propose a domain-based, integrated framework for program synthesis that emphasizes the reuse of software components and past experience in solving problems. A crucial component of the framework is a concept dictionary that contains a description of domain related components. This forms a basis for communicating with a user in a high-level, application-oriented terms; for constructing generic, problem-solving rules, and for reasoning by analogy. The other components of the framework include a library of reusable subroutines, cliches, and derivation histories of previously solved problems, a layered rulebase, a hierarchical planner, and an analogical reasoner based on Carbonell's derivational analogy paradigm.The planner uses the rulebase and the library to decompose a given problem in a top-down manner and synthesize a plan for solving it. The layered structure of the rule-base ensures that efficient plans are preferred over inefficient ones, and the changes needed to synthesize programs for a different domain are minimal. The analogical reasoner complements the role of the planner. It uses a set of heuristics to retrieve a problem from the derivation history library that is analogous to a current problem, and replays the derivation of the retrieved problem to synthesize a program for the current problem. A solved problem is stored back into the derivation history library, and may be used to solve a different problem in a future context.A prototype system, APU, based on the above framework has been implemented for the domain of UNIX programming. We describe a set of experiments designed to test the performance of APU, and provide empirical results demonstrating the feasibility and usefulness of the approach.U of I OnlyETDs are only available to UIUC Users without author permissio