Analysis and synthesis of abstract data types through generalization from examples

Abstract

The discovery of general patterns of behavior from a set of input/output examples can be a useful technique in the automated analysis and synthesis of software systems. These generalized descriptions of the behavior form a set of assertions that can be used for validation, program synthesis, program testing, and run-time monitoring. Describing the behavior is characterized as a learning process in which the set of inputs is mapped into an appropriate transform space such that general patterns can be easily characterized. The learning algorithm must choose a transform function and define a subset of the transform space which is related to equivalence classes of behavior in the original domain. An algorithm for analyzing the behavior of abstract data types is presented and several examples are given. The use of the analysis for purposes of program synthesis is also discussed.<>