Unveiling Adversarial Examples
Tim and Hadi discuss the impact of non-robust features on model accuracy, showcasing how models can be trained to recognize images based on these features. Removing non-robust features from the dataset leads to achieving robust accuracy through standard training methods.In this clip
From this podcast

Machine Learning Street Talk (MLST)
#52 - Unadversarial Examples (Hadi Salman, MIT)
Related Questions