About Me
I am Xiangyu Yin, a postdoctoral researcher at Chalmers University of Technology, working with Prof. Chih-Hong Cheng on the EU RobustifAI project. My research lies at the intersection of Adversarial Machine Learning and AI Interpretability, with a particular focus on building trustworthy and robust AI systems.
I received my Ph.D. in Computer Science from the University of Liverpool (transferred from the University of Exeter), and hold a dual undergraduate degree from the joint program between Beijing University of Posts and Telecommunications (BUPT) and Queen Mary University of London (QMUL).
News
- 2026-06 Paper ProGRank: Probe-Gradient Reranking to Defend Dense-Retriever RAG from Corpus Poisoning accepted at ECML-PKDD 2026.
- 2026-02 Paper Fragile by Design: On the Limits of Adversarial Defenses in Personalized DreamBooth Generation accepted at AAAI 2026.
- 2025-12 Paper FALCON: Fine-Grained Activation Manipulation by Contrastive Orthogonal Unalignment for Large Language Model accepted at NeurIPS 2025.
- 2025-06 Paper Toward Linearly Regularizing the Geometric Bottleneck of Linear Generalized Attention accepted at Transactions on Machine Learning Research (TMLR).
- 2025-02 Paper A Black-Box Evaluation Framework for Semantic Robustness in Bird's Eye View Detection accepted at AAAI 2025.
- 2025-01 Paper Interpreting Safety: A LLM and STPA Approach accepted at PRICAI 2025.
- 2024-09 Paper Continuous Geometry-Aware Graph Diffusion via Hyperbolic Neural PDE accepted at ECML-PKDD 2024.
- 2024-06 Paper Boosting Adversarial Training via Fisher-Rao Norm-based Regularization accepted at CVPR 2024.
- 2024-02 Paper Representation-Based Robustness in Goal-Conditioned Reinforcement Learning accepted at AAAI 2024.
- 2024-01 Paper DIMBA: Discretely Masked Black-Box Attack in Single Object Tracking accepted at Machine Learning (Springer).
- 2023-01 Paper ODE4ViTRobustness: A Tool for Understanding Adversarial Robustness of Vision Transformers accepted at Software Impacts.
- 2021-01 Paper Temple: Learning Template of Transitions for Sample Efficient Multi-task RL accepted at AAAI 2021.
- 2019-01 Paper Self-Calibrating Scene Understanding Based on Motifnet accepted at PRCV 2019.
Research Interests
- Safety and robustness of large language models (LLMs) and multimodal LLMs (MLLMs)
- Adversarial attacks and defenses in deep learning
- Interpretability and explainability of AI systems
Education
- University of Liverpool, Ph.D. in Computer Science (Dec 2022 – Jul 2025)
- University of Exeter, Ph.D. in Computer Science (Sep 2021 – Dec 2022, transferred)
- BUPT & QMUL Joint Program, B.Eng. and B.Management (Sep 2015 – Jun 2020)