PTTA: Purifying Malicious Samples for Test-Time Model Adaptation
Published in International Conference on Machine Learning (ICML), 2026
PTTA addresses the vulnerability of test-time adaptation methods to malicious samples. By purifying the incoming batch before adaptation, PTTA ensures robust model performance even under adversarial test-time conditions.
Recommended citation: Jing Ma, Hanlin Li, Xiang Xiang. (2025). "PTTA: Purifying Malicious Samples for Test-Time Model Adaptation." ICML 2025.
Download Paper