PhD student

Room CS2.33

Department of Computer Science,

University of Warwick & Kings' College London (KCL)

Email: Hanqi.Yan@warwick.ac.uk

Google ScholarLink opens in a new window, CV (2023.09)Link opens in a new window,

GitHubLink opens in a new window, TwitterLink opens in a new window

Hanqi Yan (颜寒祺) is a Ph.D. student (2020fall) under the supervision of Prof. Yulan HeLink opens in a new window and Dr. Lin GuiLink opens in a new window, also a Visiting Student at Kings' College London (KCL), doing Natural Language Processing and Machine Learning, focusing on interpretable and robust representation learning:

(1) Considering the stochastic nature of the current deep neural networks, we are able to identify a human-friendly way to understand its data-generation process.

(2) The Deep Learning model is built on certain inductive biases and sample selection biases. We propose empirical and principle methods to alleviate representation bias and learn robust representations across various testing environment.

This webpage will not be updated any longer, please refer to this new homepage [link].

---

Education and Research Experience

• 10/2022-02/2023 Visiting Student (MBZUAI-funded) at MBZUAI, advised by Dr Kun ZhangLink opens in a new window (CMU&MBZUAI), focusing on conditional generation under identifiability guarantee.
• 07/2019-10/2019 Research Assistant (PolyU-funded) advised by Prof. Wenjie LiLink opens in a new window at HKPolyU, focusing on emotion-cause extraction.
• 2017-2020 MSc. at Peking University, major in Data Science (Computer Science and Technology).
• 2013-2017 BEng. at Beihang University, affiliated with Instrumentation and Optoelectronic Engineering Department (Information Engineering).

Selected Publications

Large Language Model:

• Y. Zhou, J. Li, Y.Xiang, H.Yan, L. Gui, Y. He. Comprehensive Survey on the Interpretation and Analysis of Emergent Abilities.
• H. Yan*, Q. Zhu*, X. Wang, L. Gui, Y. He. Steer the LLMs in the self-refinement loop via unsupervised Reward.

Robust Representation Learning:

• H. Yan*, L. Kong*, L. Gui, Y. Chi, Eric. Xing, Y. He, K. Zhang. Counterfactual Generation with Identifiability GuaranteeLink opens in a new window. Neurips23, New Orleans, US & ICML23, workshop.
• H. Yan*, H. Li*, Y. Li, L. Qian, Y. He and L. Gui. Distinguishability Calibration to In-Context LearningLink opens in a new window, EACL23 (findings), Dubrovnik, Croatia.
• H. Yan, L. Gui, W. Li, and Y. He. Addressing Token Uniformity in Transformers via Singular Value TransformationLink opens in a new window. UAI22(spotlight), Eindhoven, Netherlands.
• H. Yan, L. Gui, G. Pergola and Y. He. Position Bias Mitigation: A Knowledge-Aware Graph Model for Emotion Cause ExtractionLink opens in a new window, ACL21(Oral), Remote.

Interpretability based on (statistical) Generative models:

• H. Yan*, L. Gui*, and Y. He. Hierarchical Interpretation of Neural Text ClassificationLink opens in a new window, Computational Linguistics (CL). Present in EMNLP22. Abu Dhabi, UAE
• H. Yan, L. Gui, M. Wang, K. Zhang and Y. He. Explainable Recommender with geometric information bottleneckLink opens in a new window. Under review.