Reproduce “RobustFill: Neural Program Learning under Noisy I/O” 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

Jacob Devlin, Jonathan Uesato, Surya Bhupatiraju, Rishabh Singh, Abdel-rahman Mohamed, Pushmeet Kohli

RobustFill infers executable string-transformation programs from input-output examples for spreadsheet tasks. Its neural decoder attends to examples, while search rejects candidate programs inconsistent with them. On a 205-instance benchmark, its best configuration achieved 92% all-example accuracy versus 34% for prior neural synthesis at the same beam size. This result is limited to that benchmark and does not isolate architecture effects.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.