Enhancing Gradient Inversion Attacks in Federated Learning via Hierarchical Feature Optimization

2026-04-01

AI Summary

This work addresses the privacy risks in federated learning arising from shared gradients, which can be exploited by adversaries to reconstruct users’ original data. To this end, the authors propose GIFD, a novel method that extends gradient inversion attacks from the initial latent space to the multi-level intermediate feature spaces of a generative adversarial network (GAN). By optimizing reconstruction layer-by-layer, GIFD enhances both fidelity and generalization. The approach incorporates an ℓ₁-ball constrained regularizer to suppress unrealistic artifacts and integrates a label mapping mechanism to effectively handle out-of-distribution inputs and label mismatches. Extensive experiments demonstrate that GIFD achieves high-fidelity, pixel-level data reconstruction across diverse federated learning settings, significantly outperforming existing baseline methods.

Abstract

Federated Learning (FL) has emerged as a compelling paradigm for privacy-preserving distributed machine learning, allowing multiple clients to collaboratively train a global model by transmitting locally computed gradients to a central server without exposing their private data. Nonetheless, recent studies find that the gradients exchanged in the FL system are also vulnerable to privacy leakage, e.g., an attacker can invert shared gradients to reconstruct sensitive data by leveraging pre-trained generative adversarial networks (GAN) as prior knowledge. However, existing attacks simply perform gradient inversion in the latent space of the GAN model, which limits their expression ability and generalizability. To tackle these challenges, we propose \textbf{G}radient \textbf{I}nversion over \textbf{F}eature \textbf{D}omains (GIFD), which disassembles the GAN model and searches the hierarchical features of the intermediate layers. Instead of optimizing only over the initial latent code, we progressively change the optimized layer, from the initial latent space to intermediate layers closer to the output images. In addition, we design a regularizer to avoid unreal image generation by adding a small ${l_1}$ ball constraint to the searching range. We also extend GIFD to the out-of-distribution (OOD) setting, which weakens the assumption that the training sets of GANs and FL tasks obey the same data distribution. Furthermore, we consider the challenging OOD scenario of label inconsistency and propose a label mapping technique as an effective solution. Extensive experiments demonstrate that our method can achieve pixel-level reconstruction and outperform competitive baselines across a variety of FL scenarios.

Problem

• Gradient Inversion
• Federated Learning
• Privacy Leakage
• Out-of-Distribution
• Data Reconstruction

Innovation

• gradient inversion
• federated learning
• hierarchical feature optimization
• out-of-distribution
• label mapping