Reproduce “Learning programs by learning from failures” 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
Andrew Cropper, Rolf Morel
Popper, an Inductive Logic Programming (ILP) system based on the "learning from failures" (LFF) paradigm, systematically prunes the hypothesis space by learning constraints from failed program tests. The system achieved substantial performance gains and scalability across diverse domains, outperforming state-of-the-art ILP systems in learning complex recursive programs and handling various data structures.
Key insights
- Systematic pruning of the hypothesis space via learned constraints from failures drastically improves ILP search efficiency and tractability.
- A hybrid ASP/Prolog architecture effectively combines declarative hypothesis generation with efficient runtime execution, overcoming grounding barriers and supporting complex data structures and infinite domains.
- The iterative deepening search with constraint learning guarantees the discovery of an optimal (textually minimal) logic program when one exists.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.