GP based Program Synthesis
Program Synthesis | Winter 2022 | Prof. Nadia Polikarpova
Program synthesis for well-specified programs, especially programming by example (PBE), is a well studied area with widely accepted solutions that employ constraint-based or enumerative search. Stochastic search methods like genetic programming (GP) have had success in the related domain of program repair. However, there have been surprisingly limited attempts at applying genetic programming methods to program synthesis tasks or performing comparisons on standard benchmarks with existing state-of-the-art methods. In this work, we present our implementation of a genetic programming-based synthesizer (GPSolver) for SyGuS PBE tasks. We also present a comparison with EUSolver which shows that GPSolver can outperform it in the number of benchmarks solved with the correct set of hyperparameters.
Links
Collaborators