Reproduce “Search-Aware On-Policy Self-Distillation with Sibling Alignment for Generative 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
Rongfeng Guo, Zhuang Zhuang, Ji Dai, Peng Zhao, Shijie Li, Jie Chen, Teng Long, Vincent Tao Hu and 1 more
SA-OPSD retrieves next products from interaction histories by generating semantic IDs, where beam search can prune prefixes competing for limited slots. It distills a self-teacher on states from the student’s beam search, separating target confidence from sibling preferences. On Office, its NDCG@10, a top-10 ranking metric for the held-out item, is 0.1334 versus Dense Beam’s 0.1277, while V-STAR scores 0.1340.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.