Robustness in Classification
Andrew discusses the surprising ability of classifiers to achieve high accuracy on mislabeled datasets, highlighting the existence of non-robust features that can mislead models. Yannic elaborates on the spectrum of memorization in neural networks, questioning how to differentiate between mere memorization of examples and the identification of useful features. This conversation sheds light on the complexities of training models and the implications for generalization in machine learning.In this clip
From this podcast
Machine Learning Street Talk (MLST)
Adversarial Examples and Data Modelling - Andrew Ilyas (MIT)
Related Questions