PeopleofSuperintelligence

Researcher

Junxian He

Also known as 何俊贤, 何俊賢

Junxian He is a natural language processing researcher at the Hong Kong University of Science and Technology whose work spans variational autoencoders, parameter-efficient transfer learning, benchmarks for large language models and reinforcement learning for reasoning. Shanghai Jiao Tong University gave him a bachelor's degree in electronic engineering in 2017, and Carnegie Mellon University's Language Technologies Institute gave him a PhD in 2022, with Graham Neubig and Taylor Berg-Kirkpatrick as co-advisors. Among his papers are an ICLR 2019 study of posterior collapse in variational autoencoders and a 2022 ICLR spotlight on a unified view of parameter-efficient transfer learning. The C-Eval benchmark for Chinese-language foundation models (2023) and the SimpleRL-Zoo study of zero reinforcement learning across 10 base models (2025) both list him as a coauthor. A tenure-track assistant professorship in the Department of Computer Science and Engineering at HKUST is his current post, and his research focuses on long-horizon reasoning and planning of large language models.

Badges

Timeline

  1. 2017–2022

    A PhD at Carnegie Mellon

    A bachelor's degree in Electronic Engineering from Shanghai Jiao Tong University comes in 2017. A PhD from Carnegie Mellon University's Language Technologies Institute follows in 2022, co-advised by Graham Neubig and Taylor Berg-Kirkpatrick.

  2. 2019

    Posterior collapse in VAEs

    The 2019 ICLR paper on lagging inference networks lists him as first author, with Daniel Spokoyny, Graham Neubig and Taylor Berg-Kirkpatrick. It finds that early in training the inference network of a variational autoencoder fails to approximate the model's true posterior, and it avoids posterior collapse by aggressively optimizing the inference network before each model update. A best demo paper nomination at ACL 2019 also recognizes his work.

  3. 2019–2020

    Facebook AI Research and Salesforce

    Two industry labs, Facebook AI Research in 2019 and Salesforce Research in 2020, are where he spends time.

  4. 2022

    Unified view of parameter-efficient tuning

    At ICLR 2022, the spotlight paper "Towards a Unified View of Parameter-Efficient Transfer Learning" appears with him as coauthor alongside Chunting Zhou, Xuezhe Ma, Taylor Berg-Kirkpatrick and Graham Neubig. The paper re-frames parameter-efficient methods as modifications to specific hidden states in pre-trained models and derives new methods that tune fewer parameters than earlier ones and reach results comparable to fine-tuning all parameters on four tasks. The ICLR 2022 most influential paper list from Paper Digest also includes his work.

  5. By 2023

    A faculty post at HKUST

    A seminar titled "Efficient Adaptation and Evaluation of LLMs" in November 2023 lists him as an assistant professor in the Department of Computer Science and Engineering at the Hong Kong University of Science and Technology.

  6. 2023

    The C-Eval benchmark

    C-Eval, a Chinese evaluation suite for foundation models, appears at NeurIPS 2023 with him as a coauthor. It poses multiple-choice questions across four difficulty levels and 52 disciplines, and among the models tested only GPT-4 reaches an average accuracy above 60%.

  7. 2025

    SimpleRL-Zoo and zero RL

    SimpleRL-Zoo, posted in March 2025 with him as a coauthor, studies zero reinforcement-learning training across 10 base models, including Llama3-8B, Mistral-7B/24B, DeepSeek-Math-7B and Qwen2.5 models from 0.5B to 32B; it appears at COLM 2025. The ICLR 2025 Workshop on Reasoning and Planning for Large Language Models features his keynote on the work.

  8. As of October 2026

    Tenure-track assistant professor

    A tenure-track assistant professorship in HKUST's Department of Computer Science and Engineering is his post. Long-horizon reasoning and planning of large language models is the focus of his current research.

Career

  1. NowHong Kong University of Science and TechnologyAssistant Professor
  2. 2020Salesforce
  3. 2019FAIR (Meta)

Education

  1. 2022Carnegie Mellon UniversityPhD
  2. 2017Shanghai Jiao Tong UniversityBachelor's degree in Electronic Engineering

From the biography.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Junxian He. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  2. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionIncludes Hong Kong University of Science and Technology. 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.
  3. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes Hong Kong University of Science and Technology. 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.
  4. Google has the highest combined h-index among AI organizationsIncludes Hong Kong University of Science and Technology. 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.

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 Junxian He yet. Send us a free-licensed photo.

  1. Junxian He jxhe.github.io. Checked 21 October 2026.
  2. Junxian HE | HKUST CSE cse.hkust.edu.hk. Checked 21 October 2026.
  3. Lagging Inference Networks and Posterior Collapse in Variational Autoencoders arxiv.org. Checked 21 October 2026.
  4. Towards a Unified View of Parameter-Efficient Transfer Learning arxiv.org. Checked 21 October 2026.
  5. Computer Science and Engineering Seminar - "Efficient Adaptation and Evaluation of LLMs" calendar.hkust.edu.hk. Checked 21 October 2026.
  6. C-Eval: A Multi-Level Multi-Discipline Chinese Evaluation Suite for Foundation Models arxiv.org. Checked 21 October 2026.
  7. SimpleRL-Zoo: Investigating and Taming Zero Reinforcement Learning for Open Base Models in the Wild arxiv.org. Checked 21 October 2026.
  8. ICLR Keynote Talk: Junxian He (HKUST): Taming Reinforcement Learning for Effective and Efficient Reasoners iclr.cc. Checked 21 October 2026.