Reproduce “DMRL: Document-Mediated Reinforcement Learning for Skill Optimization in Advertising Recommendation” 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
Wei Zhang, Hongji Li, Song Sun, Peng Yu, Xue Yang, Lei Zhao, Peng Jiang
DMRL uses structured skill edits and online experiments to tune advertising recommendation parameters. Its upper-level optimizer scores edits against peer rollouts and concurrent controls, penalizing disruptive changes, while a long-term reward predictor estimates seven-day outcomes from early signals. In online results, DMRL reports +0.052% LTV versus +0.024% for LRP+SkillOpt, alongside −0.020% AUD@1d; the figures lack reported significance tests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.