Reproduce “NSynC: Normalised Synthesis of Computation” 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
Zoey Shepherd, Ohad Kammar, Elizabeth Polgreen
A new program synthesis algorithm, NSynC, eliminates semantic duplication by searching directly over the unique normal forms of program semantics, demonstrating an 8.93x geometric mean speedup against a Myth* baseline on a common set of STLC+ benchmarks.
Key insights
- The space of program computations can be made semantically unique by defining and enforcing a total ordering over program normal forms, eliminating all forms of semantic duplication.
- Synthesizing programs directly as unique normal forms, derived from an extended Normalisation by Evaluation process, fundamentally improves search efficiency by ensuring every candidate represents a distinct computation.
- Strict adherence to unique normal forms, particularly with ordering constraints, can sometimes lead to increased program branching, potentially requiring more comprehensive example sets for successful synthesis.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.