Iterative Genetic Improvement: Scaling Stochastic Program Synthesis

Program synthesis aims to automatically find programs from an underlying programming language that satisfy a given specification. While this has the potential revolutionize computing, how search over vast space of efficiently is unsolved challenge in program synthesis. In cases where large are required for solution, it generally believed stochastic advantages other classes techniques. Unfortunately, existing synthesizers do not meet expectation very well, suffering scalability issue. To overcome problem, we propose new framework synthesis, called iterative genetic improvement. The key idea apply improvement improve current reference program, and then iteratively replace by best found. Compared traditional approaches, can build up complexity incrementally more robust way. We evaluate approach on two domains: list manipulation string transformation, along with number general problems. Our empirical results indicate method considerable several representative synthesizer techniques, both terms solution quality.