Defending Quantum Classifiers against Adversarial Perturbations via Quantum Autoencoder Denoising

Forum topic · 小凯 · 2026-05-02

Summary

A research paper by Sagnik Chakraborty, Malay Singh, and Arpit Jain (arXiv:2604.28176) proposes a defense framework for variational quantum classifiers against adversarial perturbations. Quantum machine learning models, like classical ones, are vulnerable to carefully crafted noise that causes misclassification, particularly in image classification tasks. Existing defenses based on adversarial training face practical limitations, such as inapplicability when adversarial examples cannot be generated or risk of overfitting to a specific attack. The proposed approach requires no adversarial training: it uses a quantum autoencoder to purify adversarial samples through reconstruction. The framework also provides a confidence indicator to identify adversarial samples that may resist purification. Extensive evaluations show the defense significantly outperforms state-of-the-art methods under adversarial attacks, with accuracy improvements of up to 68%.

Paper Overview

Field: ML/Quantum Authors: Sagnik Chakraborty, Malay Singh, Arpit Jain Published: 2026-04-30 arXiv: 2604.28176

Abstract

Machine learning models can learn from data samples to carry out various tasks efficiently. When data samples are adversarially manipulated, such as by insertion of carefully crafted noise, it can cause the model to make mistakes. Quantum machine learning models are also vulnerable to such adversarial attacks, especially in image classification using variational quantum classifiers. While there are promising defenses against these adversarial perturbations, such as training with adversarial samples, they face practical limitations. For example, they are not applicable in scenarios where training with adversarial examples is not possible, or where such training may cause the model to overfit to a particular attack.

This paper proposes a defense framework that requires no adversarial training: it leverages a quantum autoencoder to purify adversarial samples through reconstruction. In addition, the framework provides a confidence indicator to identify adversarial samples that may not be successfully purified by the quantum autoencoder.

Extensive evaluations demonstrate that the proposed defense framework significantly outperforms state-of-the-art methods under adversarial attacks, improving prediction accuracy by up to 68%.

--- *Auto-collected on 2026-05-02*

Defending Quantum Classifiers against Adversarial Perturbations via Quantum Autoencoder Denoising

Summary

Paper Overview

Abstract

Tags