DPFNAS: Differential Privacy-Enhanced Federated Neural Architecture Search for 6G Edge Intelligence

Abstract--The Sixth-Generation (6G) network envisions pervasive artificial intelligence (AI) as a core goal, enabled by edge intelligence through on-device data utilization. T o realize this vision, federated learning (FL) has emerged as a key paradigm for collaborative training across edge devices. However, the sensitivity and heterogeneity of edge data pose key challenges to FL: parameter sharing risks data reconstruction, and a unified global model struggles to adapt to diverse local distributions. In this paper, we propose a novel federated learning framework that integrates personalized differential privacy (DP) and adaptive model design. T o protect training data, we leverage sample-level representations for knowledge sharing and apply a personalized DP strategy to resist reconstruction attacks. T o ensure distribution-aware adaptation under privacy constraints, we develop a privacy-aware neural architecture search (NAS) algorithm that generates locally customized architectures and hyperparameters. T o the best of our knowledge, this is the first personalized DP solution tailored for representation-based FL with theoretical convergence guarantees. Our scheme achieves strong privacy guarantees for training data while significantly outperforming state-of-the-art methods in model performance. Experiments on benchmark datasets such as CIF AR-10 and CIF AR-100 demonstrate that our scheme improves accuracy by 6.82% over the federated NAS method PerFedRLNAS, while reducing model size to 1/10 and communication cost to 1/20. CCORDING to the International Telecommunication Union (ITU), the Sixth-Generation (6G) mobile communication network are expected to fundamentally reshape current network architectures [1]. This transformation will be driven by an unprecedented degree of connectivity. These edge devices--such as smartphones, wearables, and sensors--will continuously generate vast volumes of local data. These data, rich in contextual information and latent intelligence, are key enablers for delivering efficient and responsive artificial intelligent (AI) services. Nowadays, the utilization of data generated at the edge is still significantly limited in the Fifth-Generation mobile communication system (5GS). Y ang Lv is with the School of Cyber Engineering, Xidian University, Xi'an, China (e-mail: lyuyang@stu.xidian.edu.cn).