Multi-objective Improvement of Software Using Co-evolution and Smart Seeding
Optimising non-functional properties of software is an important part of the implementation process. One such property is execution time, and compilers target a reduction in execution time using a variety of optimisation techniques. Compiler optimisation is not always able to produce semantically equivalent alternatives that improve execution times, even if such alternatives are known to exist. Often, this is due to the local nature of such optimisations. In this paper we present a novel framework for optimising existing software using a hybrid of evolutionary optimisation techniques. Given as input the implementation of a program or function, we use Genetic Programming to evolve a new semantically equivalent version, optimised to reduce execution time subject to a given probability distribution of inputs. We employ a co-evolved population of test cases to encourage the preservation of the program’s semantics, and exploit the original program through seeding of the population in order to focus the search. We carry out experiments to identify the important factors in maximising efficiency gains. Although in this work we have optimised execution time, other non-functional criteria could be optimised in a similar manner.
KeywordsPareto Front Original Program Gain Score Test Data Generation Strength Pareto Evolutionary Algorithm
Unable to display preview. Download preview PDF.
- 2.Arcuri, A., Yao, X.: Coevolving programs and unit tests from their specification. In: IEEE International Conference on Automated Software Engineering (ASE), pp. 397–400 (2007)Google Scholar
- 3.Arcuri, A., Yao, X.: A novel co-evolutionary approach to automatic software bug fixing. In: IEEE Congress on Evolutionary Computation (CEC), pp. 162–168 (2008)Google Scholar
- 9.White, D.R., Clark, J., Jacob, J., Poulding, S.: Searching for Resource-Efficient Programs: Low-Power Pseudorandom Number Generators. In: GECCO 2008: Proceedings of the 10th Annual Conference on Genetic and Evolutionary Computation, pp. 1775–1782 (2008)Google Scholar
- 10.Zitzler, E., Laumanns, M., Thiele, L.: SPEA2: Improving the Strength Pareto Evolutionary Algorithm. Technical Report 103, Swiss Federal Institute of Technology (2001)Google Scholar
- 11.Arcuri, A., Lehre, P.K., Yao, X.: Theoretical runtime analyses of search algorithms on the test data generation for the triangle classification problem. In: International Workshop on Search-Based Software Testing (SBST), pp. 161–169 (2008)Google Scholar
- 14.ECJ: Evolutionary computation in Java, http://www.cs.gmu.edu/~eclab/projects/ecj/
- 16.Montgomery, D.C.: Design and Analysis of Experiments. John Wiley & Sons, Chichester (2006)Google Scholar