Published in submitted, 2026
Why does A beat B, B beat C, but C beat A? In standard theory, this is a paradox. In our framework, it is simply a missing dimension.
Recommended citation: Luo, Y. Fang, X., Han, R. & Xu, Y. (2026). Deep Ranking with Heterogeneous Effect. arXiv preprint arXiv:
Published in submitted, 2026
Do you follow the ATP tennis tour? Have you ever placed a bet at the race track? Who is the G.O.A.T.? Whether it’s ranking sports legends or training the latest Large Language Models (LLMs), the challenge is the same: how do we find the best when we cannot compare everyone against everyone?
Recommended citation: Fang, X., Han, R., Luo, Y. & Xu, Y. (2026). Recent Advances in the Bradley–Terry Model: Theory, Algorithms, and Applications. arXiv preprint arXiv:2601.14727.
Published in KDD2026, 2025
Architecture‑dependent, non-asymptotic regret bounds that can be sharpened under a simple margin condition (when human preferences are more “clear” than you think).
Recommended citation: Luo, Y., Ge, Y., Han, R. & Shen, G. (2025). Learning Guarantee of Reward Modeling Using Deep Neural Networks. arXiv preprint arXiv:2505.06601.
Published in submitted, 2024
Real-time learning from streaming data! We propose a one-pass, time- and memory-efficient method for high-dimensional GLMs—Adaptive Debiased Lasso (ADL)—that updates coefficients and standard errors on arrival via stochastic gradients with online debiasing.
Recommended citation: Han, R., Luo, L., Luo, Y., Lin, Y., & Huang, J. (2024). Adaptive debiased lasso in high-dimensional GLMs with streaming data. arXiv preprint arXiv:2405.18284.