Reproduce “Langevin-Informed Transfer Learning: Replacing Target Samples by Black-Box Feedback” 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
Vladimir R. Kostic, Karim Lounici, Hélène Halconruy, Timothée Devergne, Michele Parrinello, Massimiliano Pontil
The authors introduce LITL to infer slow dynamics from samples and scalar feedback, without target trajectories or gradients. It learns low-dimensional slow modes and estimates drift only within that subspace. In the Chignolin experiment, it recovers similar folding patterns from BioEmu and molecular-dynamics data, but predicts different state occupancies; this does not establish kinetics for all static generative models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.