CARSO: Blending adversarial training and representation-conditional purification via aggregation improves adversarial robustness

Code for model training, evaluation, and reproduction of the experiments contained in the paper “Blending adversarial training and representation-conditional purification via aggregation improves adversarial robustness” by Emanuele Ballarin, Alessio Ansuini and Luca Bortolussi (2025).