Reproduce “Neural Learning of One-of-Many Solutions for Combinatorial Problems in Structured Output Spaces” 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
Yatin Nandwani, Deepanshu Jindal, Mausam, Parag Singla
The paper introduces One-of-Many Learning, in which a learned selector chooses among supplied valid puzzle completions to supervise a non-autoregressive solver. The selector is trained with reinforcement learning, while the solver uses expected loss over candidate choices. On multiple-solution Sudoku tests, SELECTR reached 81.73% accuracy versus 76.93% for MINLOSS, across three runs; it selects candidates rather than generating new ones.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.