Privacy-Preserving AI Framework for 6G-Enabled Consumer Electronics

Abstract

In the realm of consumer electronics for 6G communication, AI has emerged as a significant player. However, the proliferation of devices at the edge of network causes the generation of extensive multimodal data, encompassing user behavior records, audio, and video. The influx of data poses fresh challenges concerning security and privacy. Consequently, there has been a surge in research and the implementation of AI-driven methods to protect privacy in response to these challenges. A differential privacy federated learning framework with adaptive clipping, which uses Gaussian mechanism, is proposed to mitigate privacy issue. Simultaneously, conventional federated learning depends on a centralized server and is susceptible to single points of failure and malicious node attacks. The explicit transmission of intermediate parameters can lead to the inference of private data. Therefore, a federated learning model based on blockchain is proposed to enhance decentralization, security, and fairness. Results demonstrate that the proposed framework achieves more accurate results than centralized federated learning, decentralized wireless federated learning, fused real-time sequential deep extreme learning machine, and federated learning combined with blockchain and local differential privacy, increasing the classification accuracy by 13.25%, reducing the training loss, training time, and communication overhead by 28.36%, 51.73%, and 61.44% respectively. IEEE