FedRE framework tackles federated learning's trilemma with representation entanglement

Researchers from the China Academy of Information and Communications Technology (CAICT) and Tsinghua University have introduced FedRE, a novel framework for federated learning that addresses the long-standing challenge of balancing model performance, data privacy, and communication costs. FedRE utilizes a technique called "representation entanglement" to fuse local data representations from different classes into a single entangled representation, which is then uploaded to the server. This approach not only enhances privacy by making individual data points harder to reconstruct but also reduces communication overhead by requiring clients to send only one entangled representation per round. Experiments demonstrate that FedRE achieves competitive performance, superior privacy protection, and lower communication costs compared to existing methods, particularly in model-heterogeneous federated learning scenarios. AI

影响 Offers a balanced approach to federated learning, potentially enabling more secure and efficient data collaboration in privacy-sensitive applications.

排序理由 Academic paper proposing a new method for federated learning. [lever_c_demoted from research: ic=1 ai=1.0]

在 量子位 (QbitAI) 阅读 →

• China Academy of Information and Communications Technology
• FedAllRep
• Federated Representation Entanglement
• FedGH
• FedProto
• LG-FedAvg
• Tsinghua University

• 论文
• 安全
• 产品

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →