Scholarlink

With the rapid development of artificial intelligence technology, multimodal data fusion has become a key technology for improving system performance. However, privacy leakage risks significantly increase during the integration of multi-dimensional data, and traditional differential privacy protection methods are difficult to adapt to complex multimodal environments. To address this challenge, this paper constructs an information theory-based risk quantification model. By introducing cross-modal correlation measures and dynamic privacy budget allocation strategies, precise control of privacy protection strength is achieved. An adaptive noise injection mechanism and inter-modal dependency decoupling algorithm are designed, effectively reducing data utility loss. Experimental results show that this mechanism improves data availability by 23% while ensuring strong privacy protection, providing theoretical support and technical guarantee for secure deployment of multimodal systems.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Risk Quantification and Optimization Mechanisms for Differential Privacy Protection in Multimodal Data Fusion

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jue Wang,
  • Yunlong Ge,
  • Xuan Wang

摘要

With the rapid development of artificial intelligence technology, multimodal data fusion has become a key technology for improving system performance. However, privacy leakage risks significantly increase during the integration of multi-dimensional data, and traditional differential privacy protection methods are difficult to adapt to complex multimodal environments. To address this challenge, this paper constructs an information theory-based risk quantification model. By introducing cross-modal correlation measures and dynamic privacy budget allocation strategies, precise control of privacy protection strength is achieved. An adaptive noise injection mechanism and inter-modal dependency decoupling algorithm are designed, effectively reducing data utility loss. Experimental results show that this mechanism improves data availability by 23% while ensuring strong privacy protection, providing theoretical support and technical guarantee for secure deployment of multimodal systems.