Uncovering Adversarial Risks of Test-Time Adaptation

TL;DR

This paper reveals a security vulnerability in test-time adaptation (TTA) methods, demonstrating how malicious test data can manipulate model predictions, and proposes defenses to improve TTA robustness.

Contribution

It uncovers a novel adversarial attack on TTA called Distribution Invading Attack (DIA) and evaluates countermeasures to enhance TTA security.

Findings

DIA effectively causes misclassification in TTA models across multiple benchmarks.

Countermeasures can mitigate the impact of DIA, improving TTA robustness.

TTA security vulnerabilities pose significant risks for deployment in real-world scenarios.

Abstract

Recently, test-time adaptation (TTA) has been proposed as a promising solution for addressing distribution shifts. It allows a base model to adapt to an unforeseen distribution during inference by leveraging the information from the batch of (unlabeled) test data. However, we uncover a novel security vulnerability of TTA based on the insight that predictions on benign samples can be impacted by malicious samples in the same batch. To exploit this vulnerability, we propose Distribution Invading Attack (DIA), which injects a small fraction of malicious data into the test batch. DIA causes models using TTA to misclassify benign and unperturbed test data, providing an entirely new capability for adversaries that is infeasible in canonical machine learning pipelines. Through comprehensive evaluations, we demonstrate the high effectiveness of our attack on multiple benchmarks across six TTA methods. In response, we investigate two countermeasures to robustify the existing insecure TTA implementations, following the principle of "security by design". Together, we hope our findings can make the community aware of the utility-security tradeoffs in deploying TTA and provide valuable insights for developing robust TTA approaches.