Privacy-Aware Multimodal Analysis and Its Applications

Title

Organizers

• Min Cao (mcao@suda.edu.cn)
• Mang Ye (mangye16@gmail.com )
• Xu Yang (xuyang_palm@seu.edu.cn)
• Vassili Kovalev (vassili.kovalev@gmail.com )

Abstract

The integration and analysis of multi-modal data have become a cornerstone in advancing artificial intelligence, enabling richer perception and deeper understanding by leveraging complementary information from diverse data sources. However, the increasing reliance on sensitive and personal data—particularly in domains such as surveillance, healthcare, and human-computer interaction—raises significant privacy and ethical concerns. These concerns often limit access to high-quality, multi-modal datasets, thereby hindering the development and deployment of robust intelligent systems.

This special session aims to address the critical challenges in Privacy-Aware Multimodal Analysis, focusing on methods that enable effective learning and inference while preserving individual privacy. We seek contributions that explore novel techniques and frameworks for handling data heterogeneity, bridging semantic gaps, ensuring data security, and maintaining model interpretability under privacy constraints. Additional considerations include scalability, computational efficiency, and fairness in multi-modal learning pipelines.

Our goal is to foster interdisciplinary collaboration among researchers in multimedia, computer vision, natural language processing, and machine learning to advance the state-of-the-art in privacy-preserving AI technologies. This session will highlight recent innovations and practical applications that balance performance with ethical responsibility in real-world scenarios.

Topics

• Secure and federated learning for multi-modal representation
• Privacy-preserving cross-modal transfer and adaptation
• Semi-supervised and self-supervised learning under privacy constraints
• Domain generalization and adaptation with privacy guarantees
• Adversarial training and robustness in privacy-aware multi-modal models
• Differential privacy in reinforcement learning for multimedia applications
• Privacy-preserving techniques for cross-modal retrieval systems
• Development of new benchmarks, evaluation metrics, and datasets for multi-modal research

Submission Site

https://cmt3.research.microsoft.com/mmasia2025

Code of Conduct