LGCRMLDec 7, 2022

Reconstructing Training Data from Model Gradient, Provably

arXiv:2212.03714v337 citationsh-index: 26
Originality Highly original
AI Analysis

This poses a severe threat to privacy in federated learning by exposing sensitive training data through a provable attack.

The paper tackles the problem of data privacy by showing that training samples can be fully reconstructed from a single gradient query at a random parameter value, even without training or memorization, using an efficient tensor decomposition algorithm.

Understanding when and how much a model gradient leaks information about the training sample is an important question in privacy. In this paper, we present a surprising result: even without training or memorizing the data, we can fully reconstruct the training samples from a single gradient query at a randomly chosen parameter value. We prove the identifiability of the training data under mild conditions: with shallow or deep neural networks and a wide range of activation functions. We also present a statistically and computationally efficient algorithm based on tensor decomposition to reconstruct the training data. As a provable attack that reveals sensitive training data, our findings suggest potential severe threats to privacy, especially in federated learning.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes