Determining Sequence of Image Processing Technique (IPT) to Detect Adversarial Attacks


Various adversarial attack methods pose a threat to secure machine learning models. Pre-processing-based defense against adversarial input was not adequate, and they are vulnerable to adaptive attacks. Our study proposed a dynamic pre-process-based defense technique leveraging a Genetic Algorithm that can defend against traditional adaptive attacks. We described our methodology and performed experiments using multiple datasets tested with several adversarial attacks. Our thorough empirical experiments exhibited encouraging outcomes indicating that the procedure can efficiently utilize an adversarial defense for any learning model.

Publication Title

SN Computer Science