Title: Biomedicine in the Age of AI and Foundation Models

Time: 02:00pm 

Venue: HW312, Haking Wong Building

Speaker(s): Prof. Lei Xing

Remark(s): 

Abstract

AI, driven by deep learning, has garnered significant attention in recent years and is increasingly being adopted for various applications in medical imaging and multi-omics data analysis in biomedicine. The remarkable success of AI and deep learning can be attributed to their unique ability to extract essential features from big data and make accurate inferences. This talk aims to update the audience on the latest advancements in the field of omics data analysis, including foundation models and large language models. It will also address the pitfalls of current data-driven approaches, summarize recent developments in interpretable AI, and offer perspectives on the applications of AI in multi-omics data analysis and precision oncology.

About the speaker

Prof. Lei Xing is the Jacob Haimson & Sarah S. Donaldson Professor and Director of Medical Physics Division of Radiation Oncology Department at Stanford University. He also holds affiliate faculty positions in Department of Electrical engineering, Institute for Computational and Mathematical Engineering (ICME), and Molecular Imaging Program at Stanford (MIPS). Prof. Xing obtained his PhD from the Johns Hopkins University in 1992. His research has been focused on AI, biomedical data science, medical imaging and image guided interventions, treatment planning and clinical decision-making. Prof. Xing is an author on more than 500 publications in high impact journals, an inventor on many issued and pending patents, and an investigator on numerous research grants. He is a fellow of AAPM, ASTRO, and AIMBE. He is the recipient of the 2023 Edith Quimby Lifetime Achievement Award of AAPM, which denotes outstanding scientific achievements in medical physics, influence on the professional development of others, and organizational leadership.

Title: Statistical Analysis of Large-Scare Item Response Data Under Measurement Noninvariance

Time: 02:30pm 

Venue: Room 301, Run Run Shaw Building

Speaker(s): Prof. Jing Ouyang

Remark(s): 

Abstract

International Large-Scale Assessments collect valuable data on educational quality and performance across countries, enabling education systems to share effective techniques and policies. A key analytical tool is the generalized factor model, which measures individuals’ latent traits such as skills and abilities. However, a major challenge arises from Differential Item Functioning (DIF), where different groups (e.g.,genders and countries) may have different probabilities of correctly answering the items after controlling for individual latent abilities. To address these challenges, we consider a covariate-adjusted generalized factor model and develop novel and interpretable conditions to address the identifiability issue. Based on the identifiability conditions, we propose a joint maximum likelihood estimation method and establish estimation consistency and asymptotic normality results for the covariate effects under a practical yet challenging asymptotic regime. Furthermore, we derive estimation and inference results for latent factors and the factor loadings. In a related line of work, we propose a novel estimation approach for multi-group DIF analysis that estimates the performance distributions of different groups and produces fair group rankings. The proposed method is applied to PISA 2022 data from the mathematics, science, and reading domains, providing insights into their DIF structures and performance rankings of countries.

About the speaker

Dr. Jing Ouyang is an Assistant Professor of Innovation and Information Management at the Business school of the University of Hong Kong. Prior to joining HKU, Jing received a Ph.D. in Statistics from the University of Michigan and a BSc. in Mathematics and Economics from the Hong Kong University of Science and Technology. Jing is generally interested in latent variable models, psychometrics, high-dimensional statistical inference, and statistical machine learning. Specifically, her research focuses on developing statistical theory, novel methodology and efficient computing tool for latent variable models to analyze high-dimensional and complex data, with interdisciplinary applications in large-scale educational assessments, psychological measurements, and biomedical sciences.

Title: Causal Representation Learning

Time: 10:30am 

Venue: Room 301, Run Run Shaw Building

Speaker(s): Dr. Guangyi Chen

Remark(s): 

Abstract

Traditional deep learning methods heavily rely on statistical correlations, often at the expense of generalization, robustness, and interpretability. In contrast, classical causal discovery techniques are well-suited for identifying causal relationships in structured tabular data but face significant challenges when applied to unstructured, high-dimensional inputs such as images and videos. Causal representation learning bridges this gap by uncovering the latent causal structure underlying observations. In this talk, we introduce the foundational principles of causal representation learning and its growing importance in trustworthy AI systems. Specifically, we discuss two central research questions:

1. Under what theoretical conditions can causal factors be identified from observed unstructured data?
2. How can learned causal representations improve the transferability, transparency, controllability, and attribution of AI systems in real-world applications?

About the speaker

Guangyi Chen is a postdoctoral research fellow at Carnegie Mellon University (CMU) and a research scientist at the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI). He currently co-leads the Causal Learning and Reasoning (CLeaR) Group with Prof. Kun Zhang. Prior to that, he received both his Ph.D. and B.S. degrees from Tsinghua University. His research interests include causality, representation learning, and visual understanding. A central focus of his work is to develop principled and practical methods for learning meaningful representations from visual data that support understanding, generation, and reasoning. He has published over 50 papers in top-tier machine learning and computer vision conferences, including NeurIPS, CVPR, ICLR, and so on, with several recognized as highlights or oral presentations. He also co-organized the Causal Representation Learning workshops at NeurIPS 2024 and ICDM 2024.

Title: From Statistical Physics to Privacy: Phase Transitions, Sampling and Differential Privacy

Time: 10:30am 

Venue: CB 308

Speaker(s): Prof. Jingcheng Liu

Remark(s): 

Abstract

Classical interacting particle systems studied in statistical physics are intimately connected to constraint satisfaction problems studied in computer science. Much progress has been made in the recent decade in understanding the "computational" phase transition in Gibbs sampling, where a sharp transition in computational tractability coincides precisely with the underlying physical phase transition in many models. I'll give a survey of my research along this line, and also highlight how these developments also lead to new perspectives and applications in differentially private optimizations.

About the speaker

Jingcheng Liu is an Associate Professor in the Theory Group of the School of Computer Science at Nanjing University. He is broadly interested in theoretical computer science, which includes randomized algorithms, computational phase transition, and differential privacy. Before that, he completed undergrad at SJTU (ACM Honors class) and PhD at UC Berkeley, and he was a Wally Baer and Jeri Weiss postdoctoral scholar at Caltech.

Title: The Implications of Side Bequest Motives on the Life Insurance Decisions of Retired Couples

Time: 10:30am 

Venue: Room 301, Run Run Shaw Building

Speaker(s): Prof. Ki Wai Chau

Remark(s): 

Abstract

Recent empirical evidence shows that the death of a first spouse in retired couples leads to a sharp decline in wealth, reflecting not only reduced income but also additional transfers to heirs outside the couple. Such ‘side’ bequests have significant financial consequences for a surviving spouse, but the existing literature on financial decision-making does not account for them. To fill this gap, we build a model for optimal life insurance, consumption and portfolio decisions of a retired couple, with side bequest motives. Using analytical results and numerical simulations, we show that side bequests substantially alter couples’ optimal life insurance and consumption decisions. In particular, we show that life insurance is an important tool that allows couples to balance their side bequest motive with the utility of a surviving spouse. Our model, therefore, highlights the importance of accounting for side bequests when making these decisions:

Title: The Wild Robot: A Journey Toward Long-Horizon Agentic Intelligence

Time: 11:00am 

Venue: HW312, Haking Wong Building

Speaker(s): Prof. Ivor W Tsang

Remark(s): 

Abstract

Long-horizon planning in robotic manipulation demands translating abstract goals into precise, executable actions while maintaining spatial, temporal, and physical consistency. However,
 language model-based planners often fail to handle extended task decomposition, constraint satisfaction, and adaptive recovery from errors. We present The Wild Robot, a framework for autonomous, feedback-driven reasoning that constructs and refines symbolic instruction graphs to guide code generation in robotic tasks. The system dynamically decomposes complex goals into coherent subtasks and generates executable control programs accordingly. When execution failures occur, it analyzes environmental feedback to induce and propagate new constraints, enabling targeted refinement without restarting the planning process. This structured, interpretable approach fosters resilience, adaptability, and transparency, significantly enhancing performance in long-horizon and constraint-sensitive robotic benchmarks. The Wild Robot represents a step toward truly agentic intelligence capable of robust, self-correcting decision-making in complex, real-world manipulation scenarios.

About the speaker

Professor Ivor W. Tsang is the Director of the A*STAR Centre for Frontier AI Research (CFAR) and an Adjunct Professor at the College of Computing and Data Science, NTU, Singapore. Since January 2022, he has led Singapore’s national initiative on Trustworthy Foundation Models under the National Multimodal LLM Programme. He also drives research on Agentic World Models and oversees major national initiatives such as the AI Singapore Materials Design Grand Challenge and the Maritime AI Programme. Under his leadership, CFAR has secured over S$23 million in strategic research funding, strengthening Singapore’s frontier AI ecosystem.

His research spans transfer learning, deep generative models, and big data analytics involving ultra–high-dimensional data. His influential work has earned international recognition, including the ARC Future Fellowship (2013), the ICCM Best Paper Award (2019), and recognition as the AI 2000 AAAI/IJCAI Most Influential Scholar in Australia (2020). An IEEE Fellow, he has made distinguished contributions to large-scale and transfer learning. He also serves on editorial boards of leading AI journals and top AI conference committees.

Title: Recent Advances in Game-Theoretic Feature Attributions for Kernel Methods and Gaussian Processes

Time: 02:30pm 

Venue: Room 301, Run Run Shaw Building

Speaker(s): Prof. Siu Lun Chau

Remark(s): 

Abstract

Kernel methods and Gaussian processes are powerful nonparametric learning frameworks grounded in positive definite kernels. Yet, their flexible black-box nature often comes at the cost of interpretability. This seminar presents recent advances in game-theoretic feature attribution for kernel methods and Gaussian processes, bridging cooperative game theory with kernel-based learning. I will discuss how these methods offer principled and computationally tractable attributions—reducing the exponential complexity of Shapley value estimation to polynomial time—and how they naturally extend to explain not only predictions, but also distributional discrepancies, dependency measures, and predictive uncertainty.

About the speaker

Siu Lun Chau is an Assistant Professor in Statistical Machine Learning at Nanyang Technological University, Singapore. His research focuses on understanding and addressing epistemic uncertainty in machine learning—how to represent, quantify, propagate, compare, and explain knowledge-level uncertainty in intelligent systems. Before joining NTU, he was a Postdoctoral Researcher at the CISPA Helmholtz Center for Information Security with Dr. Krikamol Muandet and obtained his DPhil in Statistics from the University of Oxford under the supervision of Prof. Dino Sejdinovic. His work has been recognised with the IJAR Young Researcher Award for contributions at the intersection of imprecise probability theory and machine learning.

Title: Deep Model Fusion

Time: 10:00am 

Venue: HW312, Haking Wong Building

Speaker(s): Prof. Dacheng Tao

Remark(s): 

Abstract

In recent years, we have witnessed a profound transformation in the learning paradigm of deep neural networks, especially in the applications of large language models and other foundation models. While conventional deep learning methodologies maintain their significance, they are now augmented by emergent model-centric approaches such as transferring knowledge, editing models, fusing models, or leveraging unlabelled data to tune models. Among these advances, deep model fusion techniques have demonstrated particular efficacy in boosting model performance, accelerating training, and mitigating the dependency on annotated datasets. Nevertheless, substantial challenges persist in the research and application of effective fusion methodologies and their scalability to large-scale foundation models. In this talk, we systematically present the recent advances in deep model fusion techniques. We provide a comprehensive taxonomical framework for categorizing existing model fusion approaches, and introduce our recent developments, including (1) weight learning-based model fusion and data-adaptive MoE upscaling, (2) subspace learning approaches to model fusion, and (3) enhanced multi-task model fusion incorporating pre- and post-finetuning to minimize representation bias between the merged model and task-specific models.

About the speaker

Prof. Dacheng Tao is the Distinguished University Professor and the Inaugural Director of the Generative AI Lab in the College of Computing and Data Science at Nanyang Technological University. He was an Australian Laureate Fellow and the founding director of the Sydney AI Centre at the University of Sydney, the inaugural director of JD Explore Academy and senior vice president at JD.com, and the chief AI scientist at UBTECH Robotics. He mainly applies statistics and mathematics to artificial intelligence, and his research is detailed in one monograph and over 300 publications. His publications have been cited over 160K times and he has an h-index 180+ in Google Scholar. He received the 2015 and 2020 Australian Eureka Prize, the 2018 IEEE ICDM Research Contributions Award, 2020 research super star by The Australian, the 2019 Diploma of The Polish Neural Network Society, and the 2021 IEEE Computer Society McCluskey Technical Achievement Award. He is a Fellow of the Australian Academy of Science, ACM and IEEE.

Title: Statistics-Powered AI

Time: 10:30am 

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

Speaker(s): Prof. Chengchun Shi

Remark(s): 

Abstract

We have definitely entered an era of generative artificial intelligence (AI), where large language models (LLMs) are increasingly reshaping our daily lives. Their impact is everywhere -- from education and academia to professional work and everyday life. In this talk, I will present two recent NeurIPS papers on statistics-powered AI, focusing on how statistical methodologies can enhance AI's performance in (1) aligning LLM's model outputs with human feedback, and (2) detecting LLM-generated content with rigorous guarantees. Open-source Python implementations are available at https://github.com/Mamba413/AdaDetectGPT and https://github.com/DRPO4LLM/DRPO4LLM.

Title: Task-Driven Image Restoration: Why, How, and What?

Time: 03:00pm 

Venue: HW312, Haking Wong Building

Speaker(s): Prof. Kyoung Mu Lee

Remark(s): 

Abstract

Image degradation is common in real-world scenarios due to factors such as transmission loss, limited camera capability, or poor shooting conditions. These issues often remove key high-frequency details, causing major performance drops in high-level vision tasks like classification, segmentation, and detection. Image restoration (IR) seeks to recover lost details in low-quality images using learned natural image priors, offering a potential solution to this problem. However, studies show that simply applying IR as a preprocessing step rarely restores the information most relevant to high-level tasks. This insight has led to Task-driven Image Restoration (TDIR), which focuses on enhancing visual quality in ways that directly benefit downstream vision tasks. In this talk, we will discuss the key challenges in TDIR and highlight several recent, efficient approaches to address them.

About the speaker

Kyoung Mu Lee (Fellow, IEEE) is currently the Editor-in-Chief (EiC) of the IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI); He is a distinguished professor at Seoul National University (SNU). He was the founding director of the Interdisciplinary Graduate Program in SNU. He is an Advisory Board Member of the Computer Vision Foundation (CVF). He was a Distinguished Lecturer of the Asia-Pacific Signal and Information Processing Association (APSIPA), from 2012 to 2013. He has received several awards, in particular, the Medal of Merit and the Scientist of Engineers of the Month Award from the Korean Government, in 2018 and 2020, respectively; the Most Influential Paper Over the Decade Award by the IAPR Machine Vision Application, in 2009; the ACCV Honorable Mention Award, in 2007; the Okawa Foundation Research Grant Award, in 2006, and the SNU Excellence in Research Award in 2020. He has also served as a General Chair for ICCV2019, ACMMM2018, and ACCV2018; and an Area Chair for CVPR, ICCV, and ECCV many times. He is the founding member and served as the President of the Korean Computer Vision Society (KCVS). Prof. Lee is a Fellow of IEEE, a member of the Korean Academy of Science and Technology (KAST) and the National Academy of Engineering of Korea (NAEK).