Zeping Yu

Zeping Yu

University of Manchester

đź“– About Me

I am a PhD researcher at the University of Manchester and the NaCTeM group, supervised by Prof. Sophia Ananiadou. Previously, I worked as an NLP researcher at Tencent Technology in Shanghai. I received my Bachelor’s and Master’s degrees from Shanghai Jiao Tong University, supervised by Prof. Gongshen Liu.

My research topic is Interpreting and Improving Large Language Models – From Internal Mechanisms to Reliable and Capable Systems. I develop tools and methods to explain, debug, and enhance LLMs/MLLMs — grounding predictions to diagnose failures and build more trustworthy systems.

I am actively seeking Research Scientist and Applied Scientist positions starting in Fall 2026. Please feel free to contact me at zepingyu@foxmail.com if you have any suitable openings!

đź“ť Research Interests

My research aims to make LLMs interpretable at the token and neuron level, and to use this understanding to design more reliable and capable systems.

a) Mechanistic Interpretability and Diagnostic Analysis of LLMs

  • VQALens: Diagnosing errors, shortcuts, and hallucinations in MLLMs. [System Demo]

  • Neuron-Level Attribution: Identifying important neurons for model diagnosis. [EMNLP 2024]

  • In-Context Learning Mechanisms: Understanding how LLMs perform in-context learning. [EMNLP 2024]

b) Mechanism-Guided Improvement of LLM Capabilities

  • Back Attention: Improving latent multi-hop reasoning in LLMs [EMNLP 2025]

  • Locate-then-Merge: Mitigating catastrophic forgetting in MLLMs [EMNLP 2025]

  • Arithmetic Reasoning via CNA: Analysis and pruning for arithmetic tasks [EMNLP 2024]

🔥 News

  • 2026.02: New survey: “Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models”.

  • 2025.08: Two papers are accepted by EMNLP 2025.

  • 2024.12: Paper lists of SAE and neuron in LLMs.

  • 2024.09: Three papers are accepted by EMNLP 2024.

  • 2024.04: Paper list of LLM interpretability.

đź“ť Publications

Mechanism-Guided Improvement of LLM Capabilities

[P1] Locate-then-Merge: Neuron-Level Parameter Fusion for Mitigating Catastrophic Forgetting in Multimodal LLMs

[P2] Back Attention: Understanding and Enhancing Multi-Hop Reasoning in Large Language Models

[P3] Understanding and Mitigating Gender Bias in LLMs via Interpretable Neuron Editing

[P4] Interpreting Arithmetic Mechanism in Large Language Models through Comparative Neuron Analysis

Interpretability and Diagnostic Analysis of LLMs

[P5] Understanding Multimodal LLMs: the Mechanistic Interpretability of LLaVA in Visual Question Answering

[P6] How do Large Language Models Learn In-Context? Query and Key Matrices of In-Context Heads are Two Towers for Metric Learning

[P7] Neuron-Level Knowledge Attribution in Large Language Models

Earlier Work

[E1] CodeCMR: Cross-modal retrieval for function-level binary source code matching

  • Zeping Yu, Wenxin Zheng, Jiaqi Wang, Qiyi Tang, Sen Nie, Shi Wu [NeurIPS 2020]

[E2] Order matters: Semantic-aware neural networks for binary code similarity detection

  • Zeping Yu*, Rui Cao* , Qiyi Tang, Sen Nie, Junzhou Huang, Shi Wu [AAAI 2020]

[E3] Adaptive User Modeling with Long and Short-Term Preferences for Personalized Recommendation

  • Zeping Yu, Jianxun Lian, Ahmad Mahmoody, Gongshen Liu, Xing Xie [IJCAI 2019]

[E4] Sliced recurrent neural networks

đź“– Educations

  • 2023.09 - 2026.09 (Expected), PhD of Computer Science, University of Manchester.
  • 2017.09 - 2020.03, Master of Computer Science, Shanghai Jiao Tong University.
  • 2013.09 - 2017.06, Bachelor of Engineering, Shanghai Jiao Tong University.