Co-evolutionary automated software correction: a proof of concept

The task of ensuring that a software artifact is correct can be a very time consuming process. To be able to say that an algorithm is correct is to say that it will produce results in accordance with its specifications for all valid input. One possible way to identify an incorrect implementation is through the use of automated testing (currently an open problem in the field of software engineering); however, actually correcting the implementation is typically a manual task for the software developer. In this thesis a system is presented which automates not only the testing but also the correction of an implementation. This is done using genetic programming methods to evolve the implementation itself and an appropriate evolutionary algorithm to evolve test cases. These two evolutionary algorithms are tied together using co-evolution such that each population plays a large role in the evolution of the other population. A prototype of the Co-evolutionary Automated Software Correction (CASC) system has been developed, which has allowed for preliminary experimentation to test the validity of the idea behind the CASC system --Abstract, page iii