1st International Workshop on Realistic Robustness and Generalization in Data Mining (RRoG-DM)

In conjunction with IEEE ICDM 2025

Overview

Robustness and generalization are ubiquitous in modern machine learning and data mining. Challenges arise in terms of robustness when models are faced with unforeseen corrupted data—be these data subject to malicious attacks, termed adversarial data, or those that are non-maliciously compromised, called commonly corrupted data. Simultaneously, there is a pressing question about generalization on how models will fare when applied to the future, standard data that may differ or evolve due to shifts in data, domain and/or categories. The concepts of robustness and generalization, although widely discussed, suffer from ambiguous definitions within the research community and can take on different meanings depending on the context. Some studies even suggest that robustness to adversarial data and the ability to generalize to unseen standard data could be inherently conflicting objectives. Adding to this, the practical measurement of robustness and generalization 'budget' is meant to capture the scope of realistic, noisy disturbances and shifts expected in actual environments. However, the research focus has largely been on exploring synthetic budgets, which does not reflect the real-world complexity (e.g. in autonomous driving or electric arc anomaly detection). This workshop aims to bring researchers and professionals from data mining and machine learning, spotlighting recent endeavors that rise to meet these challenges.

The agenda will include presentations of selected paper submissions alongside keynotes from invited academical and industrial experts. It is our ambition that through this workshop, we will not only uncover critically important avenues for the development of realistic robust and generalizable algorithms but also seed collaborative efforts that will propel this field forward.

Topics (not limited to):

• Definition and properties of realistic robustness and generalization

• Interpreting and understanding realistic model robustness and generalization

• Model uncertainty estimation and calibration for realistic robustness and generalization

*  Data approximation/augmentation/clustering/visualization with realistic robustness and generalization

*  Robust and generalization bounds and empirical evaluation criteria

*  Regularization, and the role of optimization algorithms in realistic generalization and robustness

*  Causal and logic reasoning analysis in realistic robustness and generalization

*  Counterfactual reasoning with realistic robustness and generalization

• Dataset/benchmark/ protocols for evaluating realistic robustness and generalization

• Large language models with realistic robustness and generalization

*  Architecture and devices choices that improve realistic robustness and generalization

• Applications of realistic robust and generalizable algorithms in data mining including adversarial machine learning, anomality detection/out-of-distribution detection, continuous learning, transfer learning, novel class discovery, and multi-domain generalization

*  Security-critical industrial decision making with realistic robustness and generalization

Submission:

All submissions must follow the guidelines of ICDM proceedings. The submission can be made here: ICDM RRoG Submission.

Important Dates:

• Submission: Sep 1, 2025 Sep 3, 2025

• Notification: Sep 15, 2025

• Camera-ready: Sep 25, 2025

• Workshop: Nov 12, 2025 

Organizers:
 

1) Kaizhu Huang, Professor, Duke Kunshan University, China, Email: kaizhu.huang@duke.edu
WWW: https://sites.google.com/view/kaizhu-huang-homepage
Bio: Prof. Huang’s research interests include robust learning, model generalization, and their applications in data mining and pattern recognition. He has published over 260 international journal and conference papers, e.g., in top venues IEEE TPAMI, IEEE TNNLS, ICDM, NeurIPS, AAAI, CVPR, ICCV, and ICML. He received a few best/ runner-up paper awards including 2024 IEEE ICDM 10-year Highest-impact Paper Award.

2) Zenglin Xu, Professor, Fudan University, China, Email: zenglinxu@fudan.edu.cn 
WWW: https://faculty.fudan.edu.cn/xuzenglin/en/index.htm
Bio: Prof. Xu’s main research interests include machine learning and its application in social network analysis, Internet, computational biology, and information security. He published more than 200 papers in top conferences and journals, including ICML, IJCAI, AAAI, IEEE TPAMI, IEEE TNNLS. He has repeatedly served as the committee member at major international AI conferences (e.g. AAAI/IJCAI/CVPR). He received Best Student Paper Awards in AAAI15 and ACML16.

3) Ying Gao, Jade Bird Fire (JBF), China, Email: gaoying@jbufa.com
Bio: Dr Gao is CTO in IC design, overseeing all JBF’s R&D activities. He has extensive academic and industrial experience, e.g. working at Intel Research Lab for 7 years. Dr Gao’s main research interests include reliable IC design, robust IoT design, and industrial anomality detection. He received the best paper award in 18th International Conference on Automatic Fire Detection and various industrial awards. He has been granted over 20 patents.

Advisory Chairs:

*  Weimin Cai, Jade Bird Fire, China

 *  Cheng-Lin Liu, Institute of Automation, Chinese Academy of Sciences, China

Tentative Invited Speakers:

*  Irwin King, Chinese University of Hong Kong, Hong Kong, China

* Hang Su, Tsinghua University, China

 *  Kun Zhang, Carnegie Mellon University, US

Main Contact:

Kaizhu Huang, Email: kaizhu.huang@dukekunshan.edu.cn

Mail Address: Duke Kunshan University, No. 8 Duke Avenue, Kunshan, Jiangsu, China 215316