Title: Dynamic Spectral Clustering with Provable Approximation Guarantee

Time: 10:30am 

Venue: CB308

Speaker(s): Prof. He Sun

Remark(s): 

Abstract

Spectral clustering is one of the most fundamental clustering algorithms in machine learning and has comprehensive applications in many fields of computer science. In this talk I will introduce the basics of spectral clustering, starting with its roots in spectral graph theory and its connection to eigenvalues and eigenvectors of graph Laplacians. I will present a spectral clustering algorithm in dynamic settings and discuss techniques for analyzing its performance. Several open problems will be discussed at the end of the talk. This is based on joint work with Steinar Laenen from Google Zurich, and the work appeared at ICML 2024.

About the speaker

He Sun is a Professor, and Director of Center for Algorithms and Learning Theory, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences. He received his PhD from Fudan University in 2010 and worked at the Max Planck Institute for Informatics (2010 - 2015), UC Berkeley (2014, 2023), University of Bristol (2015 - 2017), and University of Edinburgh (2017 – 2025). His research areas include algorithms, machine learning, spectral graph theory, and applied probability. He has written over 60 papers and has solved several long-standing open problems in algorithms. He received the President Medal of Fudan University (2004), Shanghai Outstanding PhD Thesis Award (2010), Simons-Berkeley Research Fellowship (2014), Turing Fellowship (2018), and EPSRC Fellowship (2020). He is a recipient of the Chinese High-Level Talent Recruitment Program for Overseas Experts (2024). He has received research grants of more than 40 million CNY, and has served as an area chair and PC member of several leading conferences in ML and TCS, including ICML 2025 and STOC 2026.

Title: HKU AI & Data Science Workshop
Date : 23 October 2025 (Thu)
Time: 2:30pm - 4:30pm
Venue: CPD 1.22

Time: 02:30pm 

Venue: CPD 1.22

Speaker(s): Prof. Xiangnan HE, Prof. Wenya WANG, Prof. Giulio CHIRIBELLA, Prof. Chao HUANG

Remark(s): 

Title: GPU Trusted Execution Environment

Time: 10:30am 

Venue: CB308

Speaker(s): Prof. Fengwei Zhang

Remark(s): 

Abstract

Trusted Execution Environments (TEEs) have been widely used for protecting endpoints and clouds for the past two decades. However, it primarily focuses on CPU processors and has not carefully considered other processors, such as GPUs. Worse, due to the vulnerable GPU software and non-confidential GPU hardware designs, attacking the GPU is not challenging and can cause severe data leakage. To address this problem, the industry/academy design GPU TEEs. We introduce two GPU TEEs: StrongBox, a GPU TEE designed for Arm endpoints such as smartphones, and CAGE, a GPU TEE tailored for Arm's latest Confidential Computing Architecture. Besides building GPU TEEs, we also discovered a GPU TEE vulnerability (MOLE) on a GPU-embedded Microcontroller Unit (MCU), which enables an attacker to leak sensitive data within the GPU TEE.

About the speaker

Dr. Fengwei Zhang is the Director of the COMPASS (COMPuter And Systems Security) Lab and a tenured Associate Professor at the Department of Computer Science and Engineering at Southern University of Science and Technology, China. Before that, he joined Wayne State University as an assistant professor at the department of computer science from 2015 to 2019. His primary research interests are in the areas of systems security, including trusted execution environments (e.g., Arm TrustZone/CCA), GPU confidential computing, debugging transparency, system introspection, and hardware- assisted security. He has published over 100 conference/journal papers, including IEEE S&P, USENIX Security, ACM CCS, NDSS, IEEE TIFS, and IEEE TDSC. He is a recipient of the Distinguished Paper Award in ACSAC 2017 and the Runner-up Best Paper Award in IEEE/IFIP DSN 2020. His high-quality work received 3 NSF Awards in the USA. He is currently the Principal Investigator of the projects from NSFC and industries. He is a senior member of ACM, a senior member of IEEE, and a distinguished member of CCF.

Title: Do Generalist Robots Need Specialist Models?

Time: 04:00pm 

Venue: CB308

Speaker(s): Prof. Chen Feng, New York University

Remark(s): 

Abstract

Large Vision-Language Models (VLMs) have demonstrated impressive generalization in the digital realm, but translating this into reliable robot manipulation and navigation remains a fundamental challenge. This talk explores a hybrid path forward: augmenting generalist "brains" with specialist "nervous systems." I will first present two foundation model efforts: SeeDo, which leverages VLMs to interpret long-horizon human videos and generate executable task plans, and INT-ACT, an evaluation suite that diagnoses a critical intention-to-execution gap in current Vision-Language-Action (VLA) systems. This gap reveals a key generalization boundary: robust task understanding does not guarantee robust physical control. To bridge this divide, I will introduce specialist models that provide two missing ingredients: fine-grained physical understanding and acquiring data for learning at scale. EgoPAT3Dv2 grounds robot action by learning 3D human intention forecasting from real-world egocentric videos. To address the data-scaling challenge, RAP employs a real-to-sim-to-real paradigm, while CityWalker explores web-scale video to learn robust, specialized skills. I will conclude by drawing analogies from the only known generalist agents—ourselves—to offer my answer to the question posed in the title.

About the speaker

Chen Feng is an Institute Associate Professor at New York University, Director of the AI4CE Lab, and Founding Co-Director of the NYU Center for Robotics and Embodied Intelligence. His research focuses on active and collaborative robot perception and robot learning to address multidisciplinary, use-inspired challenges in construction, manufacturing, and transportation. He is dedicated to developing novel algorithms and systems that enable intelligent agents to understand and interact with dynamic, unstructured environments. Prior to NYU, he worked as a research scientist in the Computer Vision Group at Mitsubishi Electric Research Laboratories (MERL) in Cambridge, Massachusetts, where he developed patented algorithms for localization, mapping, and 3D deep learning in autonomous vehicles and robotics. Chen earned his doctoral and master's degrees from the University of Michigan between 2010 and 2015, and his bachelor's degree in 2010 from Wuhan University. As an active contributor to the AI and robotics communities, he has published over 90 papers in top conferences and journals such as CVPR, ICCV, RA-L, ICRA, and IROS, and has served as an area chair and associate editor. In 2023, he was awarded the NSF CAREER Award. More information about his research can be found at https://ai4ce.github.io.

Title: HKU Innovation Week
Date: 13-14 Oct 2025
Main Event: 14 Oct 2025, 14:00-17:00, Loke Yew Hall, HKU

Time:  

Venue:  Loke Yew Hall, HKU

Speaker(s): 

Remark(s): 

HKU Innovation Week 

HKU Innovation Week 2025 marks the University’s annual celebration of innovation and entrepreneurship, highlighting the societal contributions and achievements of its students, staff, and alums. The event aims to motivate the younger HKU community to drive positive change and make a significant impact globally.

For details and registration, please visit to Https://tec.hku.hk/innovation-week/

Title: Quantum satellites and tests of relativity

Time: 10:30am 

Venue: CB308, HKU

Speaker(s): Daniel Terno

Remark(s): 

Abstracct

Quantum key distribution and other quantum 2.0 technologies are now being deployed in space. Ambitious sensitivity and stability targets at this frontier reach to previously discarded relativistic effects. Once these effects are within the sensitivity range, we are getting new tools to probe fundamental physics but also facing new fundamental limits on device performance. After briefly outlining several potentially interesting effects I will describe how combatting them leads to a true reference-frame independent QKD and describe how the technology for reliable quantum communications can be used to test of the Einstein equivalence principle.

About the speaker

Prof. Daniel Terno is a Professor at the School of Mathematical and Physical Sciences, Macquarie University Research Centre in Quantum Science and Technology, Astrophysics and Space Technologies Research Centre. He obtained his PhD at Technion in Haifa, Israel, with Asher Peres as the thesis advisor in 2003. After his PhD he moved to Perimeter Institute in Canada for a postdoctoral fellowship, and subsequently joined the faculty of the Macquarie University in Sydney in 2007. He is one of the pioneers of relativistic quantum information, which explores the connection between quantum information technologies and spacetime physics. The use of quantum systems for GPS technologies, and more broadly satellite technologies is one of the spin-offs of this fundamental area of research.

Title: From Digital Human to Humanoids

Time: 10:00am 

Venue: Rn 301, 3/F Run Run Shaw Building

Speaker(s): Prof. He Xiaodong, JD Group

Remark(s): 

Abstract

Thanks to significant technological improvements in artificial intelligence over the last decade, especially the recent development of large language and vision models that enable AI systems to communicate with humans in a very natural manner, we are now able to create high quality “digital humans” that can be deployed for various business tasks. Moreover, by integrating these models into robots and other devices, we are able to build humanoids that can listen, watch, and talk to users, serving us in a broad range of scenarios. In this talk, I will introduce our latest breakthroughs and applications in these areas and demonstrate how AI technology profoundly changes business models and user experiences while creating societal value.

About the speaker

Dr. Xiaodong He is IEEE Fellow, Senior Vice President of JD Group (JD.COM), and managing director of JD AI Research. He is the recipient of the Wu Wen-Jun AI Science & Technology Outstanding Contribution Award. He has published more than 200 papers with over 60,000 citations, and has received awards including the IEEE SPS Best Paper Award, ACL Outstanding Paper Award, and CIKM Test-of-Time Award. Since joined JD.com in 2018, he has led JD's AI effort and developed JD's JoyAI Large Language Model, and has incubated various AI products including JD's intelligent customer service and multimodal digital human Live Streaming, etc. These products and technologies have been deployed across industries including retail, logistics, finance, and healthcare. Before joining JD.com, Dr. Xiaodong He worked with Microsoft Research Redmond as Principal Researcher and Head of the Deep Learning Technology Center (DLTC). Dr. He holds a bachelor's degree from Tsinghua University and Ph.D. from the University of Missouri-Columbia. He serves as Affiliate Professor at the University of Washington (Seattle) and other institutes.

Title: Towards Efficient and Adaptive RL Algorithms in Dynamic and Multi-Agent Environments

Time: 01:30pm 

Venue: CB328

Speaker(s): Prof. Shuai Li, Shanghai Jiao Tong University

Remark(s): 

Abstract

"Reinforcement Learning (RL) in dynamic and multi-agent environments poses critical theoretical and practical challenges. These include (i) the difficulty of decomposing complex structured actions, (ii) robust convergence across diverse and non-stationary scenarios, and (iii) balancing individual learning dynamics with multi-agent game-theoretic interactions. This talk presents a unified framework addressing these challenges through (a) integrating decomposable action structures into agent learning,
(b) leveraging statistical optimization techniques for automatic adaptation to environmental variations, and (c) designing equilibrium-aware algorithms for balancing multi-agent learning and strategic interactions.
I will highlight key theoretical contributions from my group, including optimal regret bounds and robust convergence guarantees under complex function approximations. These results, published in top venues such as SODA, COLT, and ICML, have established several state-of-the-art benchmarks. I will also share successful real-world applications: achieving a bronze medal in the International SAT Competition with Huawei’s EDA parallel solver, reducing A/B testing durations by 10% in Tencent’s WeChat experimentation platform, and lowering privilege restriction rates by 69% (from 6.36 to 1.96 per 1,000) with the same level of risk control capability in Ant Group’s Risk Management Platform.
Looking forward, I will outline new directions for developing scalable RL theories and algorithms under dynamic environments with complex function approximation and multi-agent Markov games."

About the speaker

Shuai Li is an Associate Professor at the AI School of Shanghai Jiao Tong University and Deputy Director of the John Hopcroft Center for Computer Science. Her research focuses on reinforcement learning theory and algorithms for autonomous decision-making in dynamic environments, as well as the analysis of diffusion and large language models. She serves as Area Chair or Senior PC member for leading conferences such as ICML, NeurIPS, ICLR, ACL, AISTATS, IJCAI, AAMAS, and UAI. She has received the AAAI-IAAI Deployed Application Award, Shanghai Rising Talent, Shanghai Xuhui Guangqi Talent, Google PhD Fellowship, Huawei Spark Award, Tencent Outstanding Mentor Award, and international recognition in the SAT Competition. Prof. Li’s algorithms have been successfully deployed in large-scale industrial systems, significantly improving performance and efficiency.

Title: Surgical Data Science for Women's Health

Time: 04:30pm 

Venue: CB308, HKU

Speaker(s): Pierre Jannin and Prof. Krystel Nyangoh Timoh

Remark(s): 

Abstracct

Despite the growing awareness on health disparities, there still remains an important gap in research efforts between men’s and women’s health. A PubMed’s request in December 2024 on the two most common gender specific surgeries returned 22,534 results for “radical prostatectomy”, and only 3,779 for “radical hysterectomy”. There are other important women’s health issues which still have too little consideration, such as endometriosis which affects approximately 10% of women worldwide, and would benefit from research efforts in computer assisted surgery. 

Surgical data science has an important potential for improving surgical management. This may impact the whole perioperative process from diagnosis, strategy decision, planning, performance and post-operative evaluation, as well as initial and continuous learning. In the presentation, we will present the challenges related to women health and how the ones related to surgical management can be addressed by Surgical Data Science and more specifically by studying and understanding surgical skills. We will present our first results in the context of women health as well as results from other surgical specialties (e.g., neurosurgery, orthopedics) showing potentiality. Examples will cover different aspects of surgical skills from technical to non-technical ones.

About the speaker

Prof. Jannin is a INSERM Research Director at the Medical School of the University of Rennes (France). He is the head of the MediCIS research group from both UMR 1099 LTSI, Inserm research institute and University of Rennes. He has more than 30-year experience in designing and developing computer assisted surgery systems.

Prof. Timoh is Professor of Gynecology at the University of Rennes and surgeon at Rennes University Hospital (France). She is Head of the Department of Anatomy, Director of the School of Surgery in Rennes, Scientific Director of the Living Lab "Women's Health", and member of the MediCIS research group from both UMR 1099 LTSI, Inserm research institute and University of Rennes.