🤖 AI Summary
Gradient inversion attacks in federated learning typically rely on explicit priors—e.g., pre-trained generative models—and struggle to adapt to heterogeneous client data distributions.
Method: This paper introduces neural architecture search (NAS) into gradient inversion for the first time, proposing an adaptive, prior-free reconstruction framework. It employs differentiable NAS to dynamically discover optimal generator architectures, jointly optimized via gradient-matching loss, multi-scale feature reconstruction, and adversarial robustness regularization—thereby implicitly learning and adapting to scenario-specific architectural priors without requiring any pre-training.
Contribution/Results: The method achieves state-of-the-art performance under challenging conditions: high-resolution images, large batch sizes, and strong defenses (e.g., gradient clipping and differential privacy), outperforming existing approaches with an average PSNR gain of 3.2 dB. It demonstrates superior generalization, robustness, and practicality while eliminating reliance on external pre-trained models.
📝 Abstract
Gradient Inversion Attacks invert the transmitted gradients in Federated Learning (FL) systems to reconstruct the sensitive data of local clients and have raised considerable privacy concerns. A majority of gradient inversion methods rely heavily on explicit prior knowledge (e.g., a well pre-trained generative model), which is often unavailable in realistic scenarios. To alleviate this issue, researchers have proposed to leverage the implicit prior knowledge of an over-parameterized network. However, they only utilize a fixed neural architecture for all the attack settings. This would hinder the adaptive use of implicit architectural priors and consequently limit the generalizability. In this paper, we further exploit such implicit prior knowledge by proposing Gradient Inversion via Neural Architecture Search (GI-NAS), which adaptively searches the network and captures the implicit priors behind neural architectures. Extensive experiments verify that our proposed GI-NAS can achieve superior attack performance compared to state-of-the-art gradient inversion methods, even under more practical settings with high-resolution images, large-sized batches, and advanced defense strategies.