ScholarWorks@숙명여자대학교

초록

The rapid proliferation of smart consumer devices has given rise to the consumer Internet of Things (CIoT), enabling immense data collection and valuable insights for enhancing consumer experiences. However, the distributed nature of CIoT systems and the sensitivity of consumer data pose significant challenges in ensuring security, privacy, and zero trust. This paper proposes a novel framework integrating robust federated learning with a main-side blockchain architecture and zero trust principles to enable secure and privacy-preserving data sharing and collaborative learning in CIoT environments. The proposed system model consists of CIoT devices as side nodes, edge servers as main nodes, and a cloud server for global aggregation. A lightweight privacy-preserving aggregation protocol is designed based on secret sharing to protect raw data during local model updates. To enhance the robustness against Byzantine attacks and data heterogeneity, a resampling-based robust aggregation method is developed, which evaluates the cosine similarity of local updates against a reference gradient securely selected via the main-side blockchain. Experiments demonstrate that the proposed framework performs well regarding model accuracy, convergence speed, and resiliency compared with state-of-the-art methods.

키워드