Zeping Yu
Scholar

Zeping Yu

Google Scholar ID: OdpmpDsAAAAJ
University of Manchester
large language modelmechanistic interpretabilitypost-trainingreasoning
Citations & Impact
All-time
Citations
835
 
H-index
8
 
i10-index
8
 
Publications
14
 
Co-authors
8
list available
Resume (English only)
Academic Achievements
  • Publications: EMNLP 2024-a, EMNLP 2024-b, EMNLP 2024-c, EMNLP 2025-a, arXiv 2024.11, etc.; Projects: Developing and applying interpretability techniques to investigate how LLMs and MLLMs perform various tasks, designing a new module 'back attention' to enhance LLMs' latent multi-hop reasoning ability, etc.
Research Experience
  • Worked as an NLP researcher at Tencent Technology in Shanghai; Currently a PhD student at the NaCTeM group, University of Manchester.
Education
  • PhD: University of Manchester, supervised by Prof. Sophia Ananiadou; Bachelor's and Master's: Shanghai Jiao Tong University, supervised by Prof. Gongshen Liu.
Background
  • Research Interests: Understanding the inner mechanisms of LLMs and MLLMs, through mechanistic interpretability to inform the design and improve the performance of these models. Focusing on fundamental capabilities such as knowledge, arithmetic, in-context learning, and advanced capabilities like latent multi-hop reasoning, visual question answering.
Miscellany
  • Actively seeking (Applied) Research Scientist positions starting in Fall 2026.