PeopleofSuperintelligence

Researcher

Rui Lu

Also known as 卢睿

Rui Lu is a machine-learning researcher at Tsinghua University who coauthored a NeurIPS 2025 Best Paper Runner-Up on reinforcement learning and LLM reasoning. After studying in Tsinghua's Yao Class from 2016 to 2021, Lu began a PhD in the Department of Automation under Gao Huang, with papers on multitask representation learning theory, offline reinforcement learning and visual backbones. In 2024, Lu visited Princeton University as a scholar and coauthored a study of whether video generation models learn physical laws. The 2025 paper on reinforcement learning with verifiable rewards (RLVR), written with Yang Yue, Zhiqi Chen and colleagues, found that the reasoning abilities of RLVR-trained models originate from and are bounded by the base model, and it won Best Paper at the ICML 2025 AI4Math workshop. As of October 2026, Lu is a PhD student at Tsinghua and is on the job market.

Badges

Timeline

  1. 2016–2021

    Yao Class and Berkeley

    From 2016 to 2021, Lu studies in the Yao Class at Tsinghua University's Institute of Interdisciplinary Information Sciences, with Yang Yuan advising the undergraduate thesis. In 2019, they spend a summer as a visiting student at the Berkeley BAIR Center under Jitendra Malik, working on indoor 3D detection and pose estimation.

  2. 2021–2022

    A PhD under Gao Huang

    In 2021, Lu starts a PhD in the Department of Automation at Tsinghua University, advised by Gao Huang. An arXiv paper from 2021 with Gao Huang and Simon S. Du, "On the Power of Multitask Representation Learning in Linear MDP", analyzes the statistical benefit of multitask representation learning in linear Markov decision processes. From 2021 to 2022, they also work as an intern researcher at SenseTime on visual backbones, supervised by Xizhou Zhu.

  3. 2022

    General function classes at NeurIPS

    With Andrew Zhao, Simon S. Du and Gao Huang, they write "Provable General Function Class Representation Learning in Multitask Bandits and MDPs" in 2022. The paper introduces the GFUCB algorithm for learning a shared representation across contextual bandits and MDPs, and NeurIPS 2022 accepts it as a spotlight.

  4. 2023

    Offline RL and curriculum learning

    NeurIPS 2023 publishes "Understanding, Predicting and Better Resolving Q-Value Divergence in Offline-RL", which they coauthor with Yang Yue, Bingyi Kang, Shiji Song and Gao Huang. ICCV 2023 also accepts their paper on generalized curriculum learning for training visual backbones, EfficientTrain, written with Yulin Wang, Yang Yue and others.

  5. 2024

    Princeton and a video-model study

    They are a visiting scholar at Princeton University in 2024, supervised by Mengdi Wang. In November 2024, they coauthor "How Far is Video Generation from World Model: A Physical Law Perspective" with Bingyi Kang, Yang Yue, Gao Huang, Jiashi Feng and others, a study of diffusion video models trained on a 2D simulation testbed for object movement and collisions. The models mimic the closest training example when they meet new cases, and ICML 2025 takes the paper.

  6. 2025

    Probing RLVR with pass@k

    A study of reinforcement learning with verifiable rewards (RLVR) across math, coding and visual reasoning benchmarks appears on arXiv in April 2025, with Lu as third author after Yang Yue and Zhiqi Chen. Using pass@k at large k as the metric, the authors find that the reasoning abilities of RLVR-trained models originate from and are bounded by the base model, and that distillation can introduce new reasoning patterns from the teacher. The paper wins Best Paper at the ICML 2025 AI4Math workshop.

  7. 2025

    A NeurIPS runner-up award

    In November 2025, NeurIPS names their RLVR paper one of three runner-ups in its 2025 best paper awards, next to four best papers. The selection committee's reflection frames the work as a test of the widely accepted assumption that RLVR elicits genuinely new reasoning capabilities.

  8. As of October 2026

    Tsinghua PhD on the job market

    As of October 2026, they are a PhD student in the Department of Automation at Tsinghua University, advised by Gao Huang, and they are on the job market. Their research interests span machine learning, computer vision and reinforcement learning, aimed at a principled theory of deep learning's effectiveness and limitations.

Career

  1. 2024Princeton UniversityVisiting scholar
  2. 2021–2022SenseTimeIntern researcher
  3. 2019University of California, BerkeleyVisiting student

Education

  1. 2021Tsinghua UniversityUndergraduate student
  2. From 2021Tsinghua UniversityPhD

From the biography.

  1. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionIncludes University of California, Berkeley. We estimate where the people shaping AI studied and where their careers led. Massachusetts Institute of Technology leads with 467 former students, including 291 recorded doctorates.
  2. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes University of California, Berkeley. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.
  3. The median h-index is 48 at universities and 14 at frontier AI labsIncludes University of California, Berkeley. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  4. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes University of California, Berkeley. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Rui Lu yet. Send us a free-licensed photo.

  1. About Me - Rui Lu lr32768.github.io. Checked 22 October 2026.
  2. CV - Rui Lu lr32768.github.io. Checked 22 October 2026.
  3. [2106.08053] On the Power of Multitask Representation Learning in Linear MDP arxiv.org. Checked 22 October 2026.
  4. [2205.15701] Provable General Function Class Representation Learning in Multitask Bandits and MDPs arxiv.org. Checked 22 October 2026.
  5. Publications - Rui Lu lr32768.github.io. Checked 22 October 2026.
  6. [2411.02385] How Far is Video Generation from World Model: A Physical Law Perspective arxiv.org. Checked 22 October 2026.
  7. [2504.13837] Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model? arxiv.org. Checked 22 October 2026.
  8. Limit of RLVR limit-of-rlvr.github.io. Checked 22 October 2026.
  9. Announcing the NeurIPS 2025 Best Paper Awards – NeurIPS Blog blog.neurips.cc. Checked 22 October 2026.