Title: Towards Dependable Systems for Privacy-Enhancing Technologies

Time: 10:00am 

Venue: CB308, 3/F, Chow Yei Ching Building, HKU

Speaker(s): Dr. Dongwei Xiao

Remark(s): 

Abstract

"Privacy-Enhancing Technologies (PETs) are foundational for a future where data can be used without compromising privacy. While the community has largely focused on advancing the cryptographic foundations of PETs, real-world security of PETs is threatened by the very software systems designed to make them accessible, including PET-oriented compilers and frameworks.

The goal of my research is to ensure that the practical systems supporting PETs are dependable. In this talk, I will present my work on developing novel, automated techniques to systematically uncover critical vulnerabilities in the software systems of PETs. I will show two thrusts of my research: (1) automatically discovering severe logic bugs in domain-specific compilers for PETs, and (2) identifying and mitigating new, subtle security risks in PET-enhanced machine learning frameworks. The tools from this research have uncovered dozens of bugs (some with high security impact) in high-stakes PET systems and have been adopted by leading PET industry users. I will conclude by discussing my future research vision towards building provably dependable PET ecosystems."

About the speaker

Dongwei Xiao is currently a Postdoctoral Fellow at the Hong Kong University of Science and Technology, working with Prof. Shuai Wang. He earned his PhD degree from the same institution. During his PhD study, he conducted research as a visiting student at ETH Zürich with Prof. Zhendong Su. He has published papers at venues like NDSS, PLDI, and ICSE, and received an ACM SIGSOFT Distinguished Paper Award in 2023. He will join the University of Birmingham as an Assistant Professor in the Fall of 2026.

Title: Optimal Time To Sell A Stock In The Presence Of Gap, Default And Volatility Risks

Time: 04:00pm 

Venue: Room 301, Run Run Shaw Building

Speaker(s): Prof. Aleksandar Mijatovic

Remark(s): 

Abstract

Consider a small investor who holds a stock that is subject to default risk and seeks to identify the optimal time to sell the asset in the sense of minimizing the prophet's drawdown, which is the ratio of the ultimate maximum of the stock price at the time of default and the value of the stock price at the moment of sale. Assuming that default occurs at a constant rate and that at the moment of default there is a random recovery value, we solve this stochastic optimisation problem explicitly in the case the log-price of the stock prior to default is modelled by a general spectrally negative Levy process. Our results reveal a decomposition of the critical drift levels of the log-stock (at which the optimal strategy changes) into gap-risk, default-risk and volatility-risk components. Moreover, we provide an algorithm for the computation of the optimal exercise policy in terms of the Levy measure, volatility and drift parameters of the Levy process and apply this algorithm to a number of widely used models in the literature.

About the speaker

Title: Forward optimized certainty equivalent and FBSDE

Time: 02:00am 

Venue: Room 301, Run Run Shaw Building

Speaker(s): Dr. Dongwei Xiao

Remark(s): 

Abstract

We extend the notion of forward performance criteria to settings with random endowment in incomplete markets. Building on these results, we introduce and develop the novel concept of forward optimized certainty equivalent (forward OCE), which offers a genuinely dynamic valuation mechanism that accommodates progressively adaptive market model updates, stochastic risk preferences, and incoming claims with arbitrary maturities. In parallel, we develop a new methodology to analyze the emerging stochastic optimization problems by directly studying the candidate optimal control processes for both the primal and dual problems. Specifically, we derive two new systems of forward-backward stochastic differential equations (FBSDEs) and establish necessary and sufficient conditions for optimality, and various equivalences between the two problems. We provide representative examples for forward performance criteria with random endowment and forward OCE. For the case of exponential criteria, we investigate the connection between forward OCE and forward entropic risk measure. Based on joint work with Yifan Sun and Thaleia Zariphopoulou.

About the speaker

Title: Semiparametric Distribution Learning Via Quantile Regression

Time: 04:00pm 

Venue: Room 301, Run Run Shaw Building

Speaker(s): Prof. Huixia Judy Wang

Remark(s): 

Abstract

"Modern data analysis increasingly requires learning not only average trends, but also heterogeneity, uncertainty, tail behavior, and how information can be fused across heterogeneous data sources. In this talk, I will discuss how the quantile regression process
provides a flexible semiparametric approach to these problems by learning conditional distributions without imposing strong parametric assumptions on their shape.

I will highlight its role in several modern statistical problems, including multiple imputation, Bayesian inference, extreme quantile analysis, and conformal prediction, where quantile processes can help construct density-based nonconformity scores and prediction regions under complex error distributions. I will also discuss rank-based data integration motivated by the fusion
of multiple epigenetic clocks for assessing biological aging. Together, these examples illustate how quantile-based thinking can move beyond mean-centered modeling toward a richer and more robust understanding of variation, uncertainty, and individualized prediction."

About the speaker

"Huixia (Judy) Wang is the William Marsh Trustee Professor in Data Science and Chair of the Statistics Department at Rice University. She previously held faculty positions at The George Washington University and North Carolina State University and served as a Program Director at the National Science Foundation from 2018 to 2022. Her research spans statistical learning, uncertainty quantification, high-dimensional inference, quantile regression, extreme value theory and applications, spatial data analysis. She is a Fellow of the American Statistical Association and the Institute of Mathematical Statistics, an elected member of
the International Statistical Institute, and currently serves as Co-Editor of Statistica Sinica"

Title: Stochastic models based on Hawkes and marked Hawkes processes and their applications in insurance

Time: 11:00am 

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

Speaker(s): Prof. Anatoliy Swishchuk

Remark(s): 

Abstract

This talk is devoted to the study of new stochastic models for risk processes basedon Hawkes and marked Hawkes processes and their applications in insurance. We first introduce those models and outline some properties. Then we will present two applications of those models in insurance: solution of Merton optimization problem and finding ruin probabilities. Numerical examples will be presented as well.

About the speaker

Title: Toward Real-World Autonomous Learning: Adaptive Control, Safe Planning, and On-Device Foundation Models

Time: 02:00pm 

Venue: CB308, 3/F, Chow Yei Ching Building, HKU

Speaker(s): Dr. Ma Hao

Remark(s): 

Abstract

Recent progress in vision-language-action models has made embodied intelligence increasingly promising, but current robotic demonstrations still expose several system-level bottlenecks, including execution mismatch, inference latency, and limited safety integration at the planning level. In this talk, I will present my research toward autonomous learning in real-world robotics through the joint lens of control, learning, and optimization. I will first introduce a model-based online learning framework for adaptive control, with rigorous convergence guarantees and successful evaluation on a pneumatic table-tennis robot, a soft robotic system, and a heavy-duty excavator. I will then discuss constraint-aware generative planning through a diffusion-based planner for obstacle avoidance in autonomous racing, where constraints are incorporated directly into the planning process. Finally, I will present my work on efficient inference of large foundation models on edge devices under memory and compute constraints, aiming to make large-model capabilities practical for real robotic deployment. Together, these directions form a system-level framework for embodied intelligence that is adaptive, safe, and deployable, and I will conclude by discussing future opportunities in vision-based and multimodal robot learning for contact-rich manipulation.

About the speaker

Hao Ma is currently a Postdoctoral Researcher at ETH Zurich and a Scientific Researcher at the Max Planck Institute for Intelligent Systems. He received his Bachelor’s degree in Energy and Power Engineering from Jilin University in 2017, his Master’s degree in Automotive Engineering from the Technical University of Munich from 2019 to 2021, and his Doctorate in Dynamic Systems and Control from ETH Zurich from 2022 to 2025. During his Ph.D., he was also affiliated with both ETH Zurich and the Max Planck Institute for Intelligent Systems through the highly competitive Max Planck-ETH Center for Learning Systems Fellowship. His research lies at the intersection of control theory and machine learning, with a focus on enabling robots to learn autonomously in the real world. His current interests include vision-based and multimodal robot learning, contact-rich manipulation, and on-device intelligence, with an emphasis on system-level solutions for real-world robotic autonomy.

Title: Beyond LLMS: Architecting the systems backbone for semantic engines and agents

Time: 03:00pm 

Venue: HW312, Haking Wong Building, HKU

Speaker(s): Dr. Fatma Özcan

Remark(s): 

Abstract

"Large Language Models (LLMs) are redefining analysis across structured and unstructured data, leading to the emergence of two primary architectural paradigms: AI or semantic engines, and data agents. Despite distinct approaches, both architectures encounter pivotal challenges, particularly in optimizing AI operators, agentic pipelines, natural language data interfaces, and AI-powered search. Centrally, embeddings and similarity search are key building blocks. This talk first addresses optimization for semantic operators, presenting an extensive evaluation of proxy models for AI query approximation. The findings demonstrate a greater than 100x cost and latency reduction for semantic filtering (AI.IF) and significant gains for semantic ranking (AI.RANK). Next, the talk examines Filtered Vector Search (FVS), a key component for semantic search and Generative AI (GenAI) applications in modern database systems. A central insight is that optimal algorithm selection is not determined solely by distance‑metric computation costs; rather, system‑level overheads play a substantial and decisive role. Finally, the talk highlights the discovery of relevant data sources as a major bottleneck and introduces a metadata reasoner agent to address this challenge."

About the speaker

"Fatma Özcan is a Principal Engineer at Systems Research@Google. Her current research focuses on GenAI and data management, vector search, platforms and infra-structure for large-scale data analysis, and natural language interfaces to data. Dr Özcan got her PhD degree in computer science from University of Maryland, College Park, and her BSc degree in computer engineering from METU, Ankara. Before joining Google, she was a Distinguished Research Staff Member and a senior manager at IBM Almaden Research Center. She has over 24 years of experience in industrial research, and has delivered core technologies into various IBM and Google products. She is the co-author of the book ""Heterogeneous Agent Systems"", and co-author of several conference papers and patents. She is an ACM Fellow and serves on the CRA board of directors, and she is the co-chair of CRA-Industry. She received the VLDB Women in Database Research Award in 2022."

Title: Towards Trustworthy Medical Intelligence

Time: 10:30am 

Venue: Innovation Wing Two, G/F, Run Run Shaw Building

Speaker(s): Dr. Huazhu Fu

Remark(s): 

Abstract

Artificial intelligence (AI) has shown transformative potential in healthcare, particularly in medical imaging and clinical decision support. However, real-world deployment of AI systems remains hindered by two fundamental challenges: lack of trustworthiness and limited clinical usability. In this talk, I will discuss recent advances aimed at bridging these gaps. First, I will introduce methodologies for uncertainty quantification and out-of-distribution detection, enabling AI models to recognize when their predictions may be unreliable—a critical feature for patient safety. Second, I will also present GlobeReady, a training-free AI platform designed for fundus disease diagnosis that operates robustly across diverse populations and clinical environments without the need for retraining or technical intervention. Together, these efforts demonstrate a pathway toward developing AI systems that are not only technically robust but also aligned with the needs and workflows of frontline healthcare professionals.

About the speaker

Dr. Huazhu Fu is a Principal Scientist at the Institute of High Performance Computing (IHPC), Agency for Science, Technology and Research (A*STAR), Singapore. His research focuses on AI for Healthcare and Trustworthy AI. He has authored over 300 publications in leading venues, with more than 40,000 citations on Google Scholar,  H-index exceeding 90. He has been recognized as a Clarivate ‘Highly Cited Researcher’ and included in the ‘World's Top 2% Scientists’ list by Stanford. He serves as an Associate Editor for IEEE Transactions on Medical Imaging (TMI), IEEE Transactions on Neural Networks and Learning Systems (TNNLS), and IEEE Journal of Biomedical and Health Informatics (JBHI). He is a Fellow of IET.

Title: Anti-concentration inequalities for the difference of maxima of gaussian random vectors

Time: 10:30am 

Venue: Room 301, Run Run Shaw Building

Speaker(s): Prof. Shuting Shen

Remark(s): 

Abstract

We derive novel anti-concentration bounds for the difference between the maximal values of two Gaussian random vectors across various settings. Our bounds are dimension-free, scaling with the dimension of the Gaussian vectors only through the smaller expected maximum of the Gaussian subvectors. In addition, our bounds hold under the degenerate covariance structures, which previous results do not cover. In addition, we show that our conditions are sharp under the homogeneous component-wise variance setting, while we only impose some mild assumptions on the covariance structures under the heterogeneous variance setting. We apply the new anticoncentration bounds to derive the central limit theorem for the maximizers of discrete empirical processes. Finally, we back up our theoretical findings with comprehensive numerical studies.

About the speaker

Shen Shuting is an Assistant Professor of Statistics & Data Science at the National University of Singapore. Before joining NUS, she was a postdoctoral fellow at the Fuqua School of Business and the Department of Biostatistics & Bioinformatics at Duke University, jointly supervised by Dr. Alexandre Belloni and Dr. Ethan X. Fang. Prior to her postdoctoral position, she obtained her PhD in Biostatistics from Harvard University in 2023, where she was jointly supervised by Dr. Xihong Lin and Dr. Junwei Lu. She earned a B.A. and a B.S. in Mathematics (dual) from Peking University in 2018. Her research interests primarily include large-scale inference, combinatorial inference, choice model asymptotics, operations research theories, applied probability, and distributed computing.

Title: From cross-modal alignment to hierarchical sharing: statistical foundations of contrastive learning for multimodal data

Time: 02:30pm 

Venue: Room 301, Run Run Shaw Building

Speaker(s): Prof. Doudou Zhou

Remark(s): 

Abstract

"Multimodal data are increasingly common in modern biomedical and machine learning applications yet learning useful representations from heterogeneous modalities remains challenging. A central issue is that different modalities may contain complementary information, but the extent and pattern of information sharing can vary substantially across modalities. In this talk, I will present two recent works that develop statistical foundations for contrastive learning in multimodal settings. The first focuses on electronic health records and studies how structured clinical codes and unstructured clinical notes can be jointly embedded through a multimodal contrastive framework. This approach connects the contrastive objective to a pointwise mutual information matrix, yielding an interpretable and privacy-preserving algorithm based on summarylevel co-occurrence information. The second work moves beyond the conventional sharedversus-private decomposition and introduces a hierarchical framework that learns globally shared, partially shared, and modality-specific representations within a unified model. I will discuss the key modeling ideas, identifiability results, recovery guarantees, and implications for downstream prediction. Together, these works highlight how principled statistical modeling can improve both the interpretability and effectiveness of multimodal representation learning."

About the speaker

Doudou Zhou is an Assistant Professor of Statistics & Data Science at the National University of Singapore. His research lies at the intersection of statistics, machine learning, and artificial intelligence, with a focus on statistical learning theory, multimodal data integration, electronic health records, and the evaluation of large language models. He develops principled methods for learning from noisy, heterogeneous, and partially observed data, with applications in biomedicine and modern AI systems.