Reproduce “High-Dimensional Black-Box Optimization via Guidance Sampling of Diffusion Models” 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
Xiaoyan Hu, Rohit Sonker, Jalal Butt, Jiayu Chen, Egemen Kolemen, Jeff Schneider
BReaD addresses black-box optimization by using a pretrained diffusion model to generate candidates and a Gaussian process to guide sampling from queried rewards and uncertainty. It updates the guide online, without a reward-labeled training set or exact reward gradients. In image tasks, it performs comparably to exact-gradient guidance and beats an MLP reward-model baseline over evaluation rounds; the study does not measure runtime.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.