Reproduce “Exploiting Feature-invariant Scaling Laws for Efficient A/B Experimentation in Recommendation Systems” 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
Zhiliang Chen, Sammy Nammari, Jiahao Su, Xiaoqiang Luo
The authors use small-scale runs to estimate whether recommendation-model changes improve full-scale performance. They fit a power-law curve and quantify uncertainty against a known baseline. Across Wukong, HSTU, and LLM-based recommenders, they report over 40 feature changes. The method can capture a feature lowering small-scale HR@10 but raising it at full scale, but depends on the scaling form fitting the case.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.