PeopleofSuperintelligence

Researcher

Houcheng Jiang

Houcheng Jiang is a researcher on model editing and safety for large language models and a master's student at the University of Science and Technology of China. After beginning his master's degree and joining the university's Lab for Data Science in 2023, he coauthored AlphaEdit in October 2024, which won an Outstanding Paper Award at ICLR 2025. As first author, he wrote Neuron-Level Sequential Editing for ACL 2025, AnyEdit for ICML 2025, DualEdit for ICLR 2026, and a 2025 paper on contrastive weak-to-strong generalization. In February 2026, he coauthored the NExT-Guard streaming safeguard, and he continues his research at the Lab for Data Science.

Timeline

  1. October 2024

    The AlphaEdit paper

    In October 2024, he coauthors AlphaEdit with Junfeng Fang, Xiang Wang, Xiangnan He, Tat-Seng Chua and others. The method projects the perturbation of a model edit onto the null space of preserved knowledge, and it raises the performance of most locating-then-editing methods by an average of 36.7% on LLaMA3, GPT2-XL and GPT-J.

  2. 2024–2025

    Neuron-level sequential editing

    He is first author of Neuron-Level Sequential Editing (NSE), posted on arXiv in October 2024. The method optimizes the target layer's hidden states and selects neurons by activation value to limit failure and forgetting in multi-round model editing, and it appears in the ACL 2025 proceedings in Vienna in July 2025.

  3. February 2025

    AnyEdit for long-form knowledge

    In February 2025, he is first author of AnyEdit, an autoregressive method that splits long-form knowledge such as poetry, code snippets and mathematical derivations into chunks and edits the key token in each one. It outperforms strong baselines by 21.5% on benchmarks including UnKEBench, AKEW and the new EditEverything dataset, and it appears at ICML 2025.

  4. 2025

    An ICLR 2025 Outstanding Paper

    In 2025, AlphaEdit wins one of three Outstanding Paper Awards at ICLR 2025, selected from 3,704 submissions.

  5. 2025–2026

    DualEdit on backdoor editing

    In June 2025, he posts DualEdit as first author. The framework addresses safety fallback in model-editing backdoors, where an edited model starts with an affirmative prefix and later reverts to refusals, and it improves attack success by 10% and reduces the fallback rate by 11% over baselines on safety-aligned LLMs. The paper is an ICLR 2026 poster in April 2026.

  6. October 2025–July 2026

    Contrastive weak-to-strong generalization

    In October 2025, he posts Contrastive Weak-to-strong Generalization on arXiv as first author with Junfeng Fang, and revises the paper in July 2026.

  7. February 2026

    Streaming safeguards with NExT-Guard

    In February 2026, he coauthors NExT-Guard: Training-Free Streaming Safeguard without Token-Level Labels with Junfeng Fang and Nachuan Chen, introducing a training-free streaming safeguard that operates without token-level labels.

  8. Since 2023

    Master's studies at USTC

    He begins a master's degree at the University of Science and Technology of China in 2023 and joins its Lab for Data Science (LDS). He is a master's student and researcher at the university focusing on model editing and large language models.

Career

  1. Since 2023University of Science and Technology of ChinaMember

Education

  1. From 2023University of Science and Technology of ChinaMaster's

From the biography.

  1. 84% of people shaping AI with recorded undergraduate and doctoral schools changed institutions for their doctorateIncludes University of Science and Technology of China. 84% of people shaping AI with recorded undergraduate and doctoral schools changed institutions for their doctorate. Across 2,116 people with both an undergraduate degree and a doctorate recorded, we trace which undergraduate schools feed which doctoral programs, how often students stay at the same school, and how paths differ by birth-country region.
  2. 44% of AI career moves go from one university to anotherIncludes University of Science and Technology of China. We estimate how 26,439 AI career moves link universities, companies, labs, nonprofits and government research centers, with decade counts, net flows and the top paths.
  3. 11% of AI doctorate holders from 2010 to 2016 founded or led a company or lab by 10 years after graduationIncludes University of Science and Technology of China. We estimate how often AI doctorate holders founded or led a company or lab by 10 years after graduation, how waiting times compare between founders and top jobs, and which universities trained the most.
  4. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes University of Science and Technology of China. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.

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 Houcheng Jiang yet. Send us a free-licensed photo.

  1. [2410.02355] AlphaEdit: Null-Space Constrained Knowledge Editing for Language Models arxiv.org. Checked 22 October 2026.
  2. [2410.04045] Neuron-Level Sequential Editing for Large Language Models arxiv.org. Checked 22 October 2026.
  3. Neuron-Level Sequential Editing for Large Language Models - ACL Anthology aclanthology.org. Checked 22 October 2026.
  4. [2502.05628] AnyEdit: Edit Any Knowledge Encoded in Language Models arxiv.org. Checked 22 October 2026.
  5. GitHub - jianghoucheng/AnyEdit: AnyEdit: Edit Any Knowledge Encoded in Language Models, ICML 2025 github.com. Checked 22 October 2026.
  6. Announcing the Outstanding Paper Awards at ICLR 2025 – ICLR Blog blog.iclr.cc. Checked 22 October 2026.
  7. Xiang Wang | Professor, USTC xiangwang1223.github.io. Checked 22 October 2026.
  8. [2506.13285] DualEdit: Mitigating Safety Fallback in LLM Backdoor Editing via Affirmation-Refusal Regulation arxiv.org. Checked 22 October 2026.
  9. ICLR Poster DualEdit: Mitigating Safety Fallback in LLM Backdoor Editing via Affirmation-Refusal Regulation iclr.cc. Checked 22 October 2026.
  10. [2510.07884] Contrastive Weak-to-strong Generalization arxiv.org. Checked 22 October 2026.
  11. [2603.02219] NExT-Guard: Training-Free Streaming Safeguard without Token-Level Labels arxiv.org. Checked 22 October 2026.
  12. Houcheng Jiang on alphaXiv alphaxiv.org. Checked 22 October 2026.