Targeted Image Reconstruction by Sampling Pre-trained Diffusion Model

TL;DR

This paper explores how pre-trained diffusion models can be exploited to perform targeted image reconstruction attacks, revealing potential privacy risks associated with such models.

Contribution

It introduces methods for generating specific data points of a target class without prior knowledge, demonstrating a novel attack vector on diffusion models.

Findings

Successful targeted image reconstruction using diffusion models

Revealed privacy vulnerabilities in pre-trained models

Demonstrated potential for model inversion attacks

Abstract

A trained neural network model contains information on the training data. Given such a model, malicious parties can leverage the "knowledge" in this model and design ways to print out any usable information (known as model inversion attack). Therefore, it is valuable to explore the ways to conduct a such attack and demonstrate its severity. In this work, we proposed ways to generate a data point of the target class without prior knowledge of the exact target distribution by using a pre-trained diffusion model.