Exposing Vulnerabilities in Explanation for Time Series Classifiers via Dual-Target Attacks
Researchers have developed a new method called TSEF (Time Series Explanation Fooler) to expose vulnerabilities in time series classification models. This dual-target attack can manipulate both the classifier's predictions and its explanations, making it possible to achieve targeted misclassification while maintaining plausible and consistent explanations. The findings suggest that explanation stability is not a reliable indicator of a model's robustness and advocate for coupling-aware evaluation methods for trustworthy time series tasks. AI
IMPACT Highlights a critical flaw in evaluating AI model safety, potentially leading to more robust time series classification systems.