Reproduce “DeepCoder: Learning to Write Programs” 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
Matej Balog, Alexander L. Gaunt, Marc Brockschmidt, Sebastian Nowozin, Daniel Tarlow
DeepCoder ranks likely program operations with a neural network, then guides discrete search for programs matching input-output examples. The network learns from synthesized programs and examples but does not generate complete code. For short programs in a restricted language, it solved 20% of tasks in 9 seconds, versus 8,181 seconds with global function-frequency priors; this result is limited to synthetic tests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.