Reproduce “Code Building Genetic Programming” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Edward Pantridge, Lee Spector
Code Building Genetic Programming (CBGP) introduces a framework for inductive program synthesis that can integrate with arbitrary existing codebases and handle polymorphic types. The system automatically discovers available functions and types, and transcribes evolved programs into human-readable source code, achieving high solution rates on novel benchmarks and competitive performance on standard ones.
Key insights
- Representing programs as typed Directed Acyclic Graphs (DAGs) combined with dynamic reification enables type-safe composition and handling of polymorphism.
- Leveraging reflection and first-class specifications allows automatic discovery of expressions from existing codebases, reducing manual configuration requirements for integrating with arbitrary environments.
- A stack-based compilation process translates linear genomes into type-safe program DAGs, facilitating single-pass evaluation across test cases and enabling transcription into human-readable host-language source code.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.