Reproduce “Multi-Objective Generative Optimization via Expected Hypervolume Fine-Tuning” 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

Juan Guevara, Riccardo De Santi, Andreas Krause, Raul Astudillo

HVFT trains generators to produce sets of candidates with complementary objective-space trade-offs, rewarding each candidate for the hypervolume it adds relative to samples from the current model. It refreshes this marginal reward after each fine-tuning update. On molecular Geometry, HVFT reaches 4.988 versus 4.335 on 100-hypervolume, but its design-space coverage and valid fraction are lower than DIFFUSIONNFT.

Summary by alphaXiv.

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