Assessing Neural Network Robustness via Adversarial Pivotal Tuning

The robustness of image classifiers is essential to their deployment in the real world. The ability to assess this resilience to manipulations or deviations from the training data is thus crucial.