PeopleofSuperintelligence

Researcher

Jingchu Gai

Also known as 盖景初

Jingchu Gai is a machine learning researcher whose work has moved from the expressive power of graph neural networks to reinforcement learning for large language models. As a mathematics undergraduate at Peking University from 2021 to 2025, they coauthored “Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness”, which introduced homomorphism expressivity and received an Outstanding Paper Honorable Mention at ICLR 2024. A follow-up on spectral invariant GNNs appeared at ICLR 2025, and a paper on robust multi-agent reinforcement learning appeared at ICML 2025. In 2026, they published two ICML papers on reinforcement learning for language models and coauthored an LLM system that raised the certified lower bound of the Steiner ratio to 0.8559. They are a PhD student in the Machine Learning Department at Carnegie Mellon University, advised by Andrej Risteski and Aditi Raghunathan.

Timeline

  1. 2021–2025

    Mathematics at Peking University

    They study mathematics in Peking University's School of Mathematical Sciences from 2021 to 2025 and earn a B.S., advised by Liwei Wang. Their undergraduate research covers graph representation learning and GNN expressiveness.

  2. January 2024

    Homomorphism expressivity of GNNs

    They coauthor “Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness” with Bohang Zhang, Yiheng Du, Qiwei Ye, Di He and Liwei Wang, posted in January 2024. The paper introduces homomorphism expressivity, a measure of how well a graph neural network counts graphs under homomorphism.

  3. 2024

    An ICLR honorable mention

    The paper is presented as an oral at ICLR 2024, where it receives an Outstanding Paper Honorable Mention. The awards committee names 5 Outstanding Paper winners and 11 Honorable Mentions in total.

  4. 2025

    Spectral invariant GNNs at ICLR

    They publish “Homomorphism Expressivity of Spectral Invariant Graph Neural Networks” at ICLR 2025, with Yiheng Du, Bohang Zhang, Haggai Maron and Liwei Wang. The paper proves that spectral invariant GNNs can homomorphism-count exactly a class of tree-like graphs called parallel trees.

  5. 2025

    Robust multi-agent reinforcement learning

    They coauthor “Breaking the Curse of Multiagency in Robust Multi-Agent Reinforcement Learning” with Laixi Shi, Eric Mazumdar, Yuejie Chi and Adam Wierman, published at ICML 2025. The paper studies distributionally robust Markov games, where agents optimize worst-case performance when game dynamics shift within an uncertainty set.

  6. January 2026

    Language models and the Steiner ratio

    They coauthor “Towards Solving the Gilbert-Pollak Conjecture via Large Language Models” with Yisi Ke, Tianyu Huang, Yankai Shu, Di He and Liwei Wang, submitted in January 2026. The system has language models write geometric lemmas as executable code and establishes a certified lower bound of 0.8559 for the Steiner ratio, up from 0.824. The paper is published at ICML 2026.

  7. 2026

    LLM reasoning papers at ICML

    They publish two ICML 2026 papers on reinforcement learning for language models, as first author. “Differential Smoothing Mitigates Sharpening and Improves LLM Reasoning” explains why RL fine-tuning loses output diversity and improves Pass@1 and Pass@k, with gains of up to 6.7% on AIME24. “Demystifying Entropy Control in LLM RL Training” derives a dynamic scheduling method that adjusts the entropy coefficient as training evolves.

  8. Since 2025

    PhD at Carnegie Mellon

    They begin a PhD in the Machine Learning Department at Carnegie Mellon University in 2025, advised by Andrej Risteski and Aditi Raghunathan. Their current research aims at building reliable foundation models, with interests in LLM reasoning and post-training.

Career

Education

  1. 2025Peking UniversityB.S.
  2. From 2025Carnegie Mellon UniversityPhD

From the biography.

  1. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionIncludes Carnegie Mellon University. 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 Carnegie Mellon University. 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. Google has the highest combined h-index among AI organizationsIncludes Carnegie Mellon University. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  4. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Carnegie Mellon University. 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.

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 Jingchu Gai yet. Send us a free-licensed photo.

  1. Jingchu Gai gaijingchu.github.io. Checked 21 October 2026.
  2. [2401.08514] Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness arxiv.org. Checked 21 October 2026.
  3. ICLR 2024 Awards iclr.cc. Checked 21 October 2026.
  4. ICLR 2024 Outstanding Paper Awards – ICLR Blog blog.iclr.cc. Checked 21 October 2026.
  5. Homomorphism Expressivity of Spectral Invariant Graph Neural Networks proceedings.iclr.cc. Checked 21 October 2026.
  6. Breaking the Curse of Multiagency in Robust Multi-Agent Reinforcement Learning proceedings.mlr.press. Checked 21 October 2026.
  7. [2601.22365] Towards Solving the Gilbert-Pollak Conjecture via Large Language Models arxiv.org. Checked 21 October 2026.
  8. Differential Smoothing Mitigates Sharpening and Improves LLM Reasoning proceedings.mlr.press. Checked 21 October 2026.
  9. Demystifying Entropy Control in LLM RL Training: Theoretical Analysis and Dynamic Scheduling proceedings.mlr.press. Checked 21 October 2026.