Reproduce “Generative Models as Adaptive Trust Regions for Discrete Bayesian Optimisation” 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
Michael Dodds, Erik Bodin, Sebastian W. Ober, Tom Diethe, Carl Henrik Ek, Henry Moss
GenTuRBO uses a fixed pretrained generator to propose structured discrete designs near a promising candidate, while a Gaussian-process surrogate and expected improvement select proposals for evaluation. Its proposal scale adapts to search feedback. In a controlled Sokoban comparison over 20 runs and 2,000 evaluations per run, mean best score was 12.450 ± 2.438, versus 5.350 ± 1.137 for Hamming + GP/EI; this supports the proposal mechanism in that setup, not general feasibility.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.