Scholarlink

Vision-language models show remarkable capabilities in medical imaging analysis, yet their deployment in federated healthcare environments faces key challenges in privacy preservation, data heterogeneity, and adversarial robustness. We present FedDental3D-ICL, a theoretical framework for federated in-context prompt learning that enables privacy-preserving collaboration across healthcare institutions without sharing sensitive patient data or model parameters. Our framework introduces four core algorithmic contributions: Multi-Modal Prompt Space (MMPS) abstraction unifying visual and textual prompt representations across 2D and 3D medical imaging modalities; Cross-Modal Prompt Alignment (CMPA) ensuring semantic consistency through information-theoretic contrastive objectives; Hierarchical Multi-Modal Optimization (HMMO) providing theoretical convergence guarantees for non-convex federated objectives; and Byzantine-Resilient Cross-Modal Aggregation (BRCMA) with differential privacy bounds. Our theoretical analysis suggests potential convergence rates of \(O(1/\sqrt{T})\) , theoretical communication complexity bounds of \(O(K \log |P|)\) compared to traditional \(O(K \cdot d)\) , and \((\varepsilon ,\delta )\) -differential privacy guarantees with optimal composition bounds. While this work establishes comprehensive mathematical foundations, empirical validation and practical implementation remain important directions for future research.

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

Federated In-Context Prompt Selection for Multi-modal 3D Dental Imaging: A Theoretical Framework with Privacy-Preserving Guarantees

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ushashi Bhattacharjee,
  • Tirtho Roy

摘要

Vision-language models show remarkable capabilities in medical imaging analysis, yet their deployment in federated healthcare environments faces key challenges in privacy preservation, data heterogeneity, and adversarial robustness. We present FedDental3D-ICL, a theoretical framework for federated in-context prompt learning that enables privacy-preserving collaboration across healthcare institutions without sharing sensitive patient data or model parameters. Our framework introduces four core algorithmic contributions: Multi-Modal Prompt Space (MMPS) abstraction unifying visual and textual prompt representations across 2D and 3D medical imaging modalities; Cross-Modal Prompt Alignment (CMPA) ensuring semantic consistency through information-theoretic contrastive objectives; Hierarchical Multi-Modal Optimization (HMMO) providing theoretical convergence guarantees for non-convex federated objectives; and Byzantine-Resilient Cross-Modal Aggregation (BRCMA) with differential privacy bounds. Our theoretical analysis suggests potential convergence rates of \(O(1/\sqrt{T})\) , theoretical communication complexity bounds of \(O(K \log |P|)\) compared to traditional \(O(K \cdot d)\) , and \((\varepsilon ,\delta )\) -differential privacy guarantees with optimal composition bounds. While this work establishes comprehensive mathematical foundations, empirical validation and practical implementation remain important directions for future research.