Reproduce “DiMOS: Doob-Guided Inference-Time Multi-Objective Search for Scientific Design” 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

Ziqing Wang, Qijie Zhu, Weimin Wu, Zeqi Ye, Minshuo Chen, Han Liu, Kaize Ding

DiMOS guides multi-objective DNA design without fine-tuning its masked diffusion model. It uses reward-scored candidate tokens during denoising and searches promising continuations, requiring reward values rather than gradients. In soft-gate Activity plus ATAC design, joint success was 98.39 ± 0.29%, versus 55.16 ± 1.55% for SVDD, at comparable measured generation time; this is an in-silico result, not experimental validation.

Summary by alphaXiv.

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