Postdoctoral Researcher at University of Oxford
About Me
I am a postdoctoral researcher at the University of Oxford (group of Prof. David A. Clifton), developing AI solutions for healthcare.
Currently working on vision-language large models for medical image understanding (e.g., cardiologist-level echocardiogram assessment) and trustworthy machine learning for medical data privacy. Previously contributed to disentangled representation learning, over-parameterized model optimization, and multi-modal learning.
News
- 📌 [5/2026] Our paper "Mitigating Algorithmic Unfairness Arising from Forgetfulness of Medical Records in Clinical AI" is published in Nature Communications.
- 📌 [3/2025] Our paper "Denoising Reuse: Exploiting Inter-frame Motion Consistency for Efficient Video Generation" is published in IEEE TCSVT.
- 📌 [12/2024] One paper accepted at NeurIPS 2024: "Once read is enough: domain-specific pretraining-free language models with cluster-guided sparse experts for long-tail domain knowledge ".
- 📌 [12/2023] One paper accepted at NeurIPS 2023: "Train faster, perform better: modular adaptive training in over-parameterized models".
- 📌 [2/2023] One paper accepted at ICLR 2023: "Over-parameterized Model Optimization with Polyak-Łojasiewicz Condition".
Professional Affiliation
-
Postdoctoral Researcher, University of Oxford 2024.7 - Present
- Department of Engineering Science, University of Oxford
- Based at Oxford Suzhou Centre for Advanced Research (China) and University of Oxford
- Principal Investigator: Prof. David A. Clifton
- Research Focus:
- Vision-Language Models for Cardiac Disease: Developing specialized vision-language models for cardiac disease diagnosis with image understanding and reasoning ability.
- Machine Unlearning for Medical Data Privacy: Creating efficient unlearning algorithms that remove patient data from trained models without retraining.
Education
-
Ph.D. in Computer Science, Fudan University 2020.9 - 2024.6
- School of Computer Science, Fudan University
- Advisor: Prof. Li Shang
- Thesis: Analysis and Application Research on Disentangled Representations
-
M.Sc. in Management Science and Engineering, University of Chinese Academy of Sciences 2017.9 - 2020.6
- School of Engineering Science, University of Chinese Academy of Sciences
- Advisor: Prof. Jie Sui
- Thesis: Research on Deep Learning Models for Rumor Detection
-
B.Sc. in Management Science and Engineering, University of Shanghai for Science and Technology 2013.9 - 2017.6
- Business School, University of Shanghai for Science and Technology
Publications
-
[Nat. Commun. 2026] Mitigating Algorithmic Unfairness Arising from Forgetfulness of Medical Records in Clinical Artificial Intelligence.
Yixuan Chen, Anshul Thakur, Abubakar Abid Soltan, Yongyi Shen, Dongsheng Li, Mingzhi Dong, Li Shang, David A. Clifton, and Yujiang Wang.
We identify an ethical dilemma in clinical AI: protecting patients' right to be forgotten compromises honest decision-making for remaining populations. We develop a fair unlearning framework that effectively removes medical records from trained models while mitigating algorithmic bias and preserving demographic equity.
[DOI]
-
[IEEE TCSVT 2025] Denoising Reuse: Exploiting Inter-frame Motion Consistency for Efficient Video Generation.
Chenyang Wang*, Shuzhe Yan*, Yixuan Chen*, Xueyang Wang, Yujiang Wang, Mingzhi Dong, Xiaochen Yang, Dongsheng Li, Rui Zhu, David Clifton, Robert P. Dick, Qin Lv, Yingying Zhao, Fan Yang, Tun Lu, and Ning Gu. (*Equal Contribution)
A motion-aware propagation mechanism that reuses latent representations between video frames, reducing computational costs in video diffusion models while preserving visual quality and style consistency.
[DOI]
-
[NeurIPS 2024] Once Read is Enough: Domain-specific Pretraining-free Language Models with Cluster-guided Sparse Experts for Long-tail Domain Knowledge.
Fang Dong, Mengyi Chen, Jixian Zhou, Yi Shi, Yixuan Chen, Mingzhi Dong, Yujiang Wang, Dongsheng Li, Xiaochen Yang, Rui Zhu, Robert P. Dick, Qin Lv, Fan Yang, Tun Lu, Ning Gu, and Li Shang.
[NeurIPS]
-
[NeurIPS 2023] Train Faster, Perform Better: Modular Adaptive Training in Over-Parameterized Models.
Yubin Shi, Yixuan Chen, Mingzhi Dong, Xiaochen Yang, Dongsheng Li, Yujiang Wang, Robert P. Dick, Qin Lv, Fan Yang, Tun Lu, Ning Gu, and Li Shang.
-
[ICLR 2023] Over-parameterized Model Optimization with Polyak-Łojasiewicz Condition.
Yixuan Chen, Yubin Shi, Mingzhi Dong, Xiaochen Yang, Dongsheng Li, Yujiang Wang, Robert P. Dick, Qin Lv, Yingying Zhao, Fan Yang, and Li Shang.
We theoretically analyze that convergence and generalization abilities of over-parameterized models can be upper-bounded by the ratio of the Lipschitz constant and the Polyak-Łojasiewicz (PL) constant (the condition number). Based on this, we propose a structured pruning method with a gating network that dynamically detects and masks out poorly-behaved nodes during training.
-
[ICLR 2022] Recursive Disentanglement Network.
Yixuan Chen, Yubin Shi, Dongsheng Li, Yujiang Wang, Mingzhi Dong, Yingying Zhao, Robert Dick, Qin Lv, Fan Yang, and Li Shang.
We formulate the compositional disentanglement learning problem from an information-theoretic perspective and propose a recursive disentanglement network (RecurD) that propagates regulatory inductive bias recursively across the compositional feature space.
-
[WWW 2022] Cross-Modal Ambiguity Learning for Multimodal Fake News Detection.
Yixuan Chen, Dongsheng Li, Peng Zhang, Jie Sui, Qin Lv, Tun Lu, and Li Shang.
We formulate the cross-modal ambiguity learning problem from an information-theoretic perspective and propose CAFE — an ambiguity-aware multimodal fake news detection method that adaptively aggregates unimodal features and cross-modal correlations based on confidence measures.
[DOI]