Annie Qu, University of California

Title: Representation Retrieval Learning for Heterogeneous Data Integration

Abstract: In the era of big data, large-scale, multi-modal datasets are increasingly ubiquitous, offering unprecedented opportunities for predictive modeling and scientific discovery. However, these datasets often exhibit complex heterogeneity, such as covariate shift, posterior drift, and missing modalities which can hinder the accuracy of existing prediction algorithms. To address these challenges, we propose a novel Representation Retrieval (R2) framework, which integrates a representation learning module (the representer) with a sparsity-induced machine learning model (the learner). Moreover, we introduce the notion of “integrativeness” for representers, characterized by the effective data sources used in learning representers, and propose a Selective Integration Penalty (SIP) to explicitly improve the property. Theoretically, we demonstrate that the R2 framework relaxes the conventional full-sharing assumption in multi-task learning, allowing for partially shared structures, and that SIP can improve the convergence rate of the excess risk bound. Extensive simulation studies validate the empirical performance of our framework, and applications to two real-world datasets further confirm its superiority over existing approaches.

Annie Qu is Professor at the Department of Statistics and Applied Probability, University of California, Santa Barbara starting July 2025. She received her Ph.D. in Statistics from the Pennsylvania State University in 1998. Qu’s research focuses on solving fundamental issues regarding structured and unstructured large-scale data and developing cutting-edge statistical methods and theory in machine learning and algorithms for personalized medicine, text mining, recommender systems, medical imaging data, and network data analyses for complex heterogeneous data. Dr. Qu was a Data Science Founder Professor of Statistics and the Director of the Illinois Statistics Office at the University of Illinois at Urbana-Champaign during her tenure in 2008-2019, and Chancellor's Professor at UC Irvine in 2020-2025. She was a recipient of the NSF Career award from 2004 to 2009. She is a Fellow of the Institute of Mathematical Statistics (IMS), the American Statistical Association, and the American Association for the Advancement of Science. She is also a recipient of IMS Medallion Award and Lecturer in 2024. She served as Journal of the American Statistical Association Theory and Methods Co-Editor from 2023 to 2025, IMS Program Secretary from 2021 to 2027 and ASA Council of Sections of Governing Board Chair in 2025. She is the recipient of the 2025 Carver Medal of IMS.

Niansheng Tang, Yunnan University

Variational Masking Generative Model for Anomaly Detection on Incomplete Tabular Data

Building anomaly detection models solely from normal data is crucial for many real-world applications. Existing deterministic autoencoder-based methods rely on reconstruction errors for anomaly detection but lack explicit modeling of probability distribution of the normal data. This limitation inherently hinders their performance and applicability when handling incomplete data. This paper proposes a novel Variational Masking Generative Model (VMGM) for tabular data anomaly detection. Specifically, we learn the data distribution via introducing learnable masks and modeling conditional distribution of the masked features of data given the unmasked features using a conditional diffusion model. By treating these masks as latent variables and approximating their intractable posterior with a variational posterior, we derive an evidence lower bound (ELBO) for the model. The learnable masks are sampled using differentiable Gumbel-Sigmoid reparameterization. This approach enables us to derive the parametric form of the ELBO for end-to-end model optimization. For anomaly detection, we construct a detection score based on the discrepancy between generated masked data and the ground-truth masked data. Extensive experiments on six diverse tabular datasets show that VMGM outperforms several state-of-the-art methods. Specifically, it improves AUROC and AUPRC by an average of 1.34% and 1.35%, respectively, over the leading baseline for anomaly detection on both complete and incomplete data.

简介：唐年胜，教授、博导，国家杰出青年科学基金获得者、教育部“长江学者”奖励计划特聘教授、国家百千万人才工程暨有突出贡献中青年专家享受国务院政府特殊津贴，国际统计学会推选会员、国际数理统计学会会士（IMS Fellow），国家自然科学基金委数学天元基金学术领导小组成员、教育部高等学校统计学类专业教学指导委员会委员。先后主持国家杰出青年科学基金、国家自然科学基金重点项目等基金；发表学术论文170余篇（其中SCI收录134篇），在科学出版社等出版学术专著4部、译著2部、教材1部，主编出版英文书籍2部；获省部级科技奖励10项。主要研究方向包括：大数据统计学习、贝叶斯统计分析、缺失数据分析、高维数据分析、生物统计等。

Weijie Su,  University of Pennsylvania

Title: Alignment in Large Language Models: Statistical and Game-Theoretic Perspectives

Abstract: Large language models (LLMs) are predominantly aligned with human preferences through reinforcement learning from human feedback (RLHF). In this talk, we explore the theoretical foundations of LLM alignment through the intertwined lenses of statistics and game theory. First, we show how the current formulation of RLHF induces a systematic bias we call preference collapse, and how this can be mitigated by introducing a tailored regularization term into the reward function. Next, we expose a fundamental bottleneck of reward-based alignment, demonstrating that cyclic human preferences cannot be faithfully represented by scalar reward models such as the Bradley-Terry model. More precisely, we establish that such cyclic inconsistencies give rise to a lower bound on the approximation error of any scalar reward fitting. Shifting to a game-theoretic perspective, we focus on Nash learning from human feedback and establish several social choice desiderata for this approach to alignment, including the preservation of preference diversity through the emergence of mixed strategies. Finally, we show that the zero-sum game approach generally cannot perfectly match a target preference distribution as a unique Nash equilibrium.

Weijie Su is an Associate Professor in the Wharton Statistics and Data Science Department and, by courtesy, in the Departments of Computer and Information Science, and Biostatistics, Epidemiology and Informatics, at the University of Pennsylvania. He serves as a co-director of the Penn Research in Machine Learning (PRiML) Center. Prior to joining Penn, he earned his PhD in from Stanford University in 2016 and his bachelor's degree from Peking University in 2011. His research interests include statistical foundations of generative AI, privacy-preserving data analysis, high-dimensional statistics, and optimization. He serves as a co-editor of Statistical Learning and Data Science (SLADS) and an associate editor of Journal of the American Statistical Association, Journal of Machine Learning Research, Annals of Applied Statistics, Operations Research, Harvard Data Science Review, and Foundations and Trends in Statistics. He is currently on the ICML 2026 Organizing Committee as the Scientific Integrity Chair.

The 2026 ICSA China Conference will be held at Southern University of Science and Technology, with the main conference taking place from June 27 to June 29. It is recommended that the participants may make room reservation. The organizing committee has recommended hotels with negotiated rates, but please note that these rates may not necessarily be the lowest. We encourage all participants to compare prices on other platforms to make the most favorable choice for themselves. For hotel information. Please refer to the table below for hotel information.

Special reminder: When booking and checking into hotels, please remind the service personnel that we are here to attend the 2026 ICSA China Conference to avoid missing the negotiation price. Please make reservations as early as possible, preferably by phone. Hotel Reservation Deadline: June 18, 2026

Hotel Information:

2026国际泛华统计协会中国会议将在南方科技大学举行，主会议将于6月27日至6月29日召开。建议参会者提前预订房间。组委会已推荐了协商价格的酒店，但请注意，这些价格可能并非最低。我们鼓励所有参会者比较其他平台的价格，以做出对自己最有利的选择。有关酒店信息，请参阅下表。

特别提醒：参会人员预订和入住酒店时，请提示服务人员是来参加2026国际泛华统计协会中国会议的，以免错失协议价。请与会人员尽早预定，最好通过电话预定。酒店留房截止日期：2026年6月18日

酒店信息：