Abstract
Large, overparameterized models such as neural networks are now the workhorses of modern machine learning. These models are often trained to near-zero error on noisy datasets and simultaneously generalize well to unseen data, in contrast to the textbook intuition regarding the perils of overfitting. Classical theoretical frameworks provide little guidance for navigating these questions due to overparameterization. It is thus crucial to develop new intuition regarding overfitting and generalization that reflect these empirical observations. In this talk, we discuss recent work in the statistical literature that provides theoretical insights into these phenomena on linear models.