Machine learning
Our machine learning theme drives the creation of intelligent algorithms that learn from data and power smarter decisions.
Spanning areas such as computer vision, explainable artificial intelligence (AI), deep learning, reinforcement learning, and multimodal large language models, this theme explores applications from healthcare and finance to robotics and sustainability. The focus is on creating trustworthy, transparent, and impactful machine learning solutions that tackle real-world challenges across society and industry.
Our research
Our existing research spans over the following few topics:
Computer Vision empowers machines to interpret, analyze, and act on visual data—transforming raw pixels or voxels into semantic understanding. We develop advanced systems that extract spatial, contextual, and relational insights from images, videos, and multi-sensor inputs, with a strong focus on applications for autonomous driving vehicles, Earth Observation, and robotic systems. In parallel, we leverage generative models to synthesize highly realistic images and videos that closely mimic real-world scenarios. This addresses critical challenges such as the scarcity of corner-case data and the difficulty of acquiring data for rare or hazardous situations. At the same time, the proliferation of AI-generated high-quality visual content on social media platforms has led to widespread issues of misinformation/disinformation, as well as fraudulent activities in sectors like insurance and finance. To combat this, we are building large-scale, authentic datasets and developing systematic detection frameworks with explainable AI to identify synthetic media. Additionally, we are actively researching the downstream impacts and real-world applications of Large Language Models (LLMs) and Vision-Language Models (VLMs), driving innovation in AI-powered solutions.
Reinforcement Learning explores how intelligent agents learn to make decisions through experience in complex environments. We develop both single-agent and multi-agent systems where agents coordinate or compete, apply RL to advance games and robotics, and create safer, more robust algorithms for real-world applications. We also leverage RL to improve large language models and agents and accelerate scientific discovery and often involve humans in the learning process. Our projects address long-term challenges in games, language reasoning, and robotics —pushing AI towards safer, more adaptable, and widely beneficial systems.
AI Safety addresses the critical issue of ensuring that AI systems behave reliably, predictably, and in alignment with human values—even in complex or unforeseen situations. This field explores how to design, train, and deploy AI models in ways that prevent harmful behaviours, reduce risks from unintended consequences, and maintain control over increasingly autonomous systems. One of our key contributions to AI Safety is on the promotion of methods that provide provable guarantees to behaviours of AI and generative AI (GenAI). These methods include various formal and practical verification methods, methods that provide statistical guarantees, coverage guided testing methods, runtime monitoring, and AI assurance. These methods, contributing to areas such as robustness to adversarial inputs, interpretability, value alignment, safe reinforcement learning, and fail-safe mechanism, support the construction of safer AI and GenAI systems, especially in high-stakes domains like healthcare, autonomous vehicles, and decision-making infrastructure.
Our research on Large Language Models (LLMs) addresses some of the most pressing challenges in their development and deployment, focusing on deepfakes, reasoning, and safety & security. We investigate how LLMs can both generate and detect synthetic content, developing robust methods to identify and mitigate the risks of deepfakes in text, audio, and multimodal contexts. We study LLM reasoning capabilities to understand their strengths, limitations, and failure modes, aiming to design models that can reason more reliably and transparently. In parallel, we advance LLM safety and security by creating safeguards against misuse, hardening models against adversarial attacks, and ensuring that outputs align with ethical and legal standards. Together, these efforts aim to make LLMs not only more capable but also more trustworthy and resilient in real-world applications.
Our work in Explainable AI (XAI) focuses on creating methods that make the decision-making processes of complex models transparent and interpretable. A contribution is BayLIME, a Bayesian extension of the popular LIME framework, which quantifies uncertainty in local explanations and provides more robust, stable insights into model behavior. Beyond BayLIME, we develop and apply a range of XAI techniques—including feature attribution, counterfactual reasoning, and example-based explanations—to uncover how models arrive at their predictions. These tools empower researchers, practitioners, and stakeholders to trust, validate, and improve AI systems, ensuring that transparency is not an afterthought but a core design principle.
People
- Prof Xiaowei Huang
AI Safety and Security, Verification, Trustworthy AI, Formal Methods, Explainable AI - Dr Guangliang Cheng
Computer Vision, Deepfake Detection, Autonomous Driving, Robotics - Dr Meng Fang
Natural Language Processing, Reinforcement Learning, Agents, Artificial intelligence - Dr Yi Dong
Deep Reinforcement Learning, Probabilistic Verification, Multi-Agent Systems, Distributed Optimisation, Power System Management - Dr Keiller Nogueira
Computer Vision, Artificial Intelligence, Earth Observation, Remote Sensing
Partnerships and collaborations
Collaborators
- The Chinese University of Hong Kong
- University of Cambridge
- Imperial College London
- University of Leeds
- Peking University
- Tsinghua University
- Institute of Automation
- Chinese Academy of Sciences
- Nanyang Technological University
- National University of Singapore, etc.
Grants
- An Ethical and Robust AI Development Framework: Assessing Correctness and Detecting Fakes. 03/2024 – 10/2025. £260K. Team: Guangliang Cheng (PI), Xiaowei Huang. Funded by Alan Turing Institute
- RobustifAI: Robustifying Generative AI through Human-Centric Integration of Neural and Symbolic Methods. 06/2025 – 05/2028. €9.3M. Team: Xiaowei Huang (PI), Guangliang Cheng, Yi Dong, Gabriella Pizzuto. Funded by the EU Commission under Horizon Europe programme
- AI-PASSPORT: Development of AI-Based Digital Platform and Service to Enhance Efficiency and Safety for Ships and PORTs. 11/2024-10/2026. £1.9M. Team: Yi Dong, Xiaowei Huang. Funded by Innovate UK under UK-South Korea CR&D 2024.
Recent publications (2023-2025 in top conferences)
- Tianxiao Li, Zhenglin Huang, Haiquan Wen, Yiwei He, Shuchang Lyu, Baoyuan Wu, Guangliang Cheng, RAIDX: A Retrieval-Augmented Generation and GRPO Reinforcement Learning Framework for Explainable Deepfake Detection. MM 2025
- Zhenglin Huang, Jinwei Hu, Xiangtai Li, Yiwei He, Xingyu Zhao, Bei Peng, Baoyuan Wu, Xiaowei Huang, Guangliang Cheng. #. CVPR 2025
- Weiguang Zhao, Rui Zhang, Qiufeng Wang, Guangliang Cheng, Kaizhu Huang. BFANet: Revisiting 3D Semantic Segmentation with Boundary Feature Analysis. CVPR 2025
- Zheng Zhou, Wenquan Feng, Qiaosheng Zhang, Shuchang Lyu, Qi Zhao, Guangliang Cheng. ROME is Forged in Adversity: Robust Distilled Datasets via Information Bottleneck. ICML 2025
- Jianan Ye, Weiguang Zhao, Xi Yang, Guangliang Cheng, Kaizhu Huang. PO3AD: Predicting Point Offsets toward Better 3D Point Cloud Anomaly Detection. CVPR 2025
- Jianan Ye, Zhaorui Tan, Yijie Hu, Xi Yang, Guangliang Cheng, Kaizhu Huang. Disentangling Tabular Data towards Better One-Class Anomaly Detection. AAAI 2025
- Xiangtai Li, Shilin Xu, Yibo Yang, Haobo Yuan, Guangliang Cheng, Yunhai Tong, Zhouchen Lin, Ming-Hsuan Yang, Dacheng Tao. Panopticpartformer++: A unified and decoupled view for panoptic part segmentation. T-PAMI 2024
- Xiangtai Li, Henghui Ding, Haobo Yuan, Wenwei Zhang, Jiangmiao Pang, Guangliang Cheng, Kai Chen, Ziwei Liu, Chen Change Loy. Transformer-based visual segmentation: A survey. T-PAMI 2024
- Xiangtai Li, Jiangning Zhang, Yibo Yang, Guangliang Cheng, Kuiyuan Yang, Yunhai Tong, Dacheng Tao. Sfnet: Faster and accurate semantic segmentation via semantic flow. IJCV 2024
- Xiangtai Li, Haobo Yuan, Wenwei Zhang, Guangliang Cheng, Jiangmiao Pang, Chen Change Loy. Tube-link: A flexible cross tube baseline for universal video segmentation. ICCV 2023
- Jianzong Wu, Xiangtai Li, Henghui Ding, Xia Li, Guangliang Cheng, Yunhai Tong, Chen Change Loy. Betrayed by captions: Joint caption grounding and generation for open vocabulary instance segmentation. ICCV 2023
- Hugo Oliveira, Caio Silva, Gabriel Machado, Keiller Nogueira, Jefersson dos Santos (2023). Fully convolutional open set segmentation. Machine Learning, 112(5), 1733-1784
- Keiller Nogueira, Mayara Faita-Pinheiro, Ana Paula Ramos, Wesley Gonçalves, Jose Marcato Junior, Jefersson dos Santos (2024). Prototypical contrastive network for imbalanced aerial image segmentation. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (pp. 8366-8376)
- Ribana Roscher, Marc Russwurm, Caroline Gevaert, Michael Kampffmeyer, Jefersson dos Santos, Maria Vakalopoulou, Ronny Hansch, Stine Hansen, Keiller Nogueira, Jonathan Prexl, Devis Tuia (2024). Better, not just more: Data-centric machine learning for earth observation. IEEE Geoscience and Remote Sensing Magazine
- D Yi, R Mu, G Jin, Y Qi, J Hu, X Zhao, J Meng, W Ruan, X Huang. Position: Building Guardrails for Large Language Models Requires Systematic Design. ICML 2024
- Yucheng Yang, Tianyi Zhou, Mykola Pechenizkiy, Meng Fang. Preference Controllable Reinforcement Learning with Advanced Multi-Objective Optimization. International Conference on Machine Learning (ICML). 2025
- Zijing Shi, Meng Fang, Ling Chen. Monte Carlo Planning with Large Language Model for Text-Based Games. International Conference on Learning Representations (ICLR). 2025
- Yudi Zhang, Pei Xiao, Lu Wang, Chaoyun Zhang, Meng Fang, Yali Du, Yevgeniy Puzyrev, Randolph Yao, Si Qin, Qingwei Lin, Mykola Pechenizkiy, Dongmei Zhang, Saravan Rajmohan, Qi Zhang. RuAG: Learned-rule-augmented Generation for Large Language Models. International Conference on Learning Representations (ICLR). 2025
- Hongye Cao, Fan Feng, Meng Fang, Shaokang Dong, Tianpei Yang, Jing Huo, Yang Gao. Towards Empowerment Gain through Causal Structure Learning in Model-Based RL. International Conference on Learning Representations (ICLR). 2025
- Jiawei Xu, Rui Yang, Shuang Qiu, Feng Luo, Meng Fang, Baoxiang Wang, Lei Han. Tackling Data Corruption in Offline Reinforcement Learning via Sequence Modeling. International Conference on Learning Representations (ICLR). 2025
- Tristan Tomilin, Meng Fang, Mykola Pechenizkiy. HASARD: A Benchmark for Harnessing Safe Reinforcement Learning with Doom. International Conference on Learning Representations (ICLR). 2025
- Xiong-Hui Chen, Ziyan Wang, Yali Du, Shengyi Jiang, Meng Fang, Yang Yu, Jun Wang. Understanding, Rehearsing, and Introspecting: Learn a Policy from Textual Tutorial Books in Football Games. Conference on Neural Information Processing Systems (NeurIPS). 2024
- Xuanfa Jin, Ziyan Wang, Yali Du, Meng Fang, Haifeng Zhang, Jun Wang. Learning to Discuss Strategically: A Case Study on One Night Ultimate Werewolf. Conference on Neural Information Processing Systems (NeurIPS). 2024
- Yucheng Yang, Tianyi Zhou, Qiang He, Lei Han, Mykola Pechenizkiy, Meng Fang. Task Adaptation from Skills: Information Geometry, Disentanglement, and New Objectives for Unsupervised Reinforcement Learning. International Conference on Learning Representations (ICLR). 2024
- Qiang He, Tianyi Zhou, Meng Fang, Setareh Maghsudi. Adaptive Regularization of Representation Rank as an Implicit Constraint of Bellman Equation. International Conference on Learning Representations (ICLR). 2024
- Yudi Zhang, Yali Du, Biwei Huang, Ziyan Wang, Jun Wang, Meng Fang, Mykola Pechenizkiy. Interpretable Reward Redistribution in Reinforcement Learning: A Causal Approach. Conference on Noeural Information Processing Systems (NeurIPS). 2023
- Tristan Tomilin, Meng Fang, Yudi Zhang, Mykola Pechenizkiy. COOM: A Game Benchmark for Continual Reinforcement Learning. Conference on Neural Information Processing Systems (NeurIPS). 2023
- Lu Yin, Gen Li, Meng Fang, Li Shen, Tianjin Huang, Zhangyang Wang, Vlado Menkovski, Xiaolong Ma, Mykola Pechenizkiy, Shiwei Liu. Dynamic Sparsity Is Channel-Level Sparsity Learner. Conference on Neural Information Processing Systems (NeurIPS). 2023
- Zijing Shi, Meng Fang, Yunqiu Xu, Ling Chen, Yali Du. Stay Moral and Explore: Learn to Behave Morally in Text-based Games. International Conference on Learning Representations (ICLR). 2023.