Hanqi Yan is a lecturer (Assistant Professor) at King’s College London, Department of Informatics, affliated with the NLP group. Before that, she was the postdoctoral researcher in the same department (2024-2025). She mainly focuses on interpretability and robustness for language models on various reasoning tasks, especially from the representation learning perspective. She has published more than 10+ papers as (co-)first author on related topics in top conferences, such as ACL, EMNLP, Neurips, ICML. She serves as area chair for ACL2025, EMNLP2025, and co-organized the student workshop at AACL2022. She obtained her Ph.D from the University of Warwick(2024), advised by Prof. Yulan He and Dr. Lin Gui, Msc and BEng from Peking University (2020) and Beihang University (2017).

I’ve been incredibly lucky to have a number of amazing collaborators and mentors across KCL and a range of other institutions, including Carnegie Mellon University, MIT, MBZUAI, Hong Kong Polytechnic University,University of Warwick. None of the research so far—would be possible without their kind help and support.

I am always on the lookout for excellent PhD students to work with me about LLM reasoning, robustenss, interpretability etc.

• Fully-funded positions for exceptional candidates through KCL scholarships. These funding allocations are for full fees and a living stipend (London weighted) for any nationality.
• K-CSC (for Chinese nationals). These funding allocations are for full fees and a living stipend (London weighted).
• Alternative fundings.

🔥 News

👩‍🏫 Professional Service

• Organiser
  • PreTrain 2025: Spotlight on ACL/ICML/ICLR at KCL
  • Co-Chair of the AACL22-Student Research Workshop
• Area Chair:
  • ACL25, EMNLP25
• Reviewers:
  • NLP: AACL, NAACL, EACL, EMNLP, ACL, COLM25
  • AI/ML: UAI, AISTATS, NEURIPS, ICLR, ICML, AAAI26
  • Journal: NeuroComputing, TOIS, TMLR, Transactions on Big Data, Transactions on Artificial Intelligence.

💬 Invited Talks

• 08/2025. Thomson Reuters, Foundational Research Team invited by Jonathan Richard Schwarz.Towards More Robust Reasoning for LLMs
• 07/2024. Fudan University, NLP Group . Representation Learning and Mechanistic Interpretability
• 03/2023. Turing AI Fellowship Event , London, Distinguishability Calibration to In-Context Learning
• 04/2022. UKRI Fellows Workshop , University of Edinburgh, Interpreting Long Documents and Recommendation Systems via Latent Variable Models
• 07/2021. SPAAM Seminar. University of Warwick. Emotion Cause detection.

🔍 Research Summary

My research interests lie in the intersection of Machine Learning and Natural Language Processing, i.e., incorporating fundamental representation learning to enhance the interpretability and reliability of different NLP models, with 10+ (co-)first-authored papers published at top-tier venues:

• Mechanistic interpretability (neuron-level) in language models and multi-modal models [EMNLP24,NeurIPS24-RBMF], explaining the conflicts between safety and reasoning advancement [Preprint]; self-explainable models with a conceptualised layer linking the input and decision layer [Computational Linguistics 22,TKDE24].
• Empirical and principled methods to enhance model robustness over various test inputs, e.g., position bias [ACL24-findings,ACL21-oral], distribution shifts [NeurIPS23] and representation inefficiency in transformer-based models [EMNLP24,EACL23-findings,UAI22-spotlight].
• Understanding and enhancing LM’s reasoning capabilities via injecting external knowledge[ACL21-oral][ACL25], weak supervision [EMNLP24], applying a self-refinement mechanism for factual knowledge reasoning [ACL24]. More recently focus in the two directions:
  • Science literature understanding, such as code generation for scientific paper replication on our own SciReplicate-Bench and novelty assessment.
  • Reasoning in latent space, such as a position paper at ICML25 about meta-reasoning, CODI in implicit CoT, in retrieved-based QA [ACL25], navigating search in latent space [ICML25spotlight] and sparse feature for preference optimization [ICML25].

📝 Publications

(* indicates equal contribution)

📝 Notes

o1-technical report (notes for video)
Machine Unlearning via CausalLens and in NLP tasks
Reading List For Large Language Model
Identifiability101 in Causality (3rd PhD)
Induction Head_ contribute to In-context Learning (3rd PhD)
Recommendation with Causality (2nd PhD)
Causality101 (Feb 2022, 2nd PhD)
Explaining Neural Networks (Oct 2020 1st PhD)