Gradient Inversion Transcript: Leveraging Robust Generative Priors to Reconstruct Training Data from Gradient Leakage

2025/05/26 by Xinping Chen, Chen Liu, Chen, Xinping +1
Computer Science · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning and Algorithms

paper · pdf · doi:10.48550/arxiv.2505.20026

openalex publication_date 2025/05/26 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

We propose Gradient Inversion Transcript (GIT), a novel generative approach for reconstructing training data from leaked gradients. GIT employs a generative attack model, whose architecture is tailored to align with the structure of the leaked model based on theoretical analysis. Once trained offline, GIT can be deployed efficiently and only relies on the leaked gradients to reconstruct the input data, rendering it applicable under various distributed learning environments. When used as a prior for other iterative optimization-based methods, GIT not only accelerates convergence but also enhances the overall reconstruction quality. GIT consistently outperforms existing methods across multiple datasets and demonstrates strong robustness under challenging conditions, including inaccurate gradients, data distribution shifts and discrepancies in model parameters.

Citations

• Recovering Labels from Local Updates in Federated Learning
• SPEAR:Exact Gradient Inversion of Batches in Federated Learning
• GIFD: A Generative Gradient Inversion Method with Feature Domain Optimization
• Reconstructing Training Data from Model Gradient, Provably
• Learning to Invert: Simple Adaptive Attacks for Gradient Inversion in Federated Learning
• Reconstructing Training Data from Trained Neural Networks
• Auditing Privacy Defenses in Federated Learning via Generative Gradient Leakage
• Gradient Inversion with Generative Image Prior
• See through Gradients: Image Batch Recovery via GradInversion
• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• R-GAP: Recursive Gradient Attack on Privacy
• SAPAG: A Self-Adaptive Privacy Attack From Gradients
• A Framework for Evaluating Gradient Leakage Attacks in Federated Learning
• Inverting Gradients -- How easy is it to break privacy in federated learning?
• iDLG: Improved Deep Leakage from Gradients
• Deep Leakage from Gradients
• Attention Is All You Need
• Deep Residual Learning for Image Recognition
• The perceptron: A probabilistic model for information storage and organization in the brain.
• Gradient-based learning applied to document recognition

Related

• Uncovering Gradient Inversion Risks in Practical Language Model Training
• GI-NAS: Boosting Gradient Inversion Attacks Through Adaptive Neural Architecture Search
• Uncovering Privacy Vulnerabilities through Analytical Gradient Inversion Attacks
• Beyond Gradients: Exploiting Adversarial Priors in Model Inversion Attacks
• Exploring User-level Gradient Inversion with a Diffusion Prior