Reproduce “In-Context Batch Black-box Optimization via Flow Matching” 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

Quoc-Anh Hoang Nguyen, Xinyu Zhang, Daolang Huang, Julien Martinelli, Samuel Kaski

FMTR addresses batch black-box design by learning a reusable sampler offline. Given observed designs and outcomes plus remaining rounds, it generates variable-size batches from a continuous proposal without test-time surrogate refitting or acquisition optimization. Its parameters remain fixed during deployment. In q=64 synthetic tests it has best average rank across tested dimensions, while results on data-derived benchmarks vary by task.

Summary by alphaXiv.

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