Adversarial Robustness in Traffic Sign Recognition

Worked on adversarial machine learning by implementing and testing two attack methods—Fast Gradient Sign Method (FGSM) and Jacobian-based Saliency Map Attack (JSMA)—against a convolutional neural network (CNN) trained on traffic sign images. Assessed how adversarial perturbations affect model accuracy and explored basic defense approaches to improve robustness in safety-critical areas like autonomous driving.