Sparse Representations Improve Adversarial Robustness of Neural Network Classifiers Paper โข 2509.21130 โข Published Sep 25