3 papers
stat.ML2026
Overfitting of Spectral Gradient Descent: How Matrix Geometry shapes Generalization and Implicit Bias
Guillaume Braun, Ichiro Hashimoto, Masaaki Imaizumi
We study the generalization of spectral gradient descent (SpecGD) in overparameterized matrix classification with corrupted labels. Each input combines a shared low-rank signal wit…
cs.LG2025
Directional Convergence, Benign Overfitting of Gradient Descent in leaky ReLU two-layer Neural Networks
Ichiro Hashimoto
In this paper, we provide sufficient conditions of benign overfitting of fixed width leaky ReLU two-layer neural network classifiers trained on mixture data via gradient descent. O…
cs.LG2025
Universality of Benign Overfitting in Binary Linear Classification
Ichiro Hashimoto, Stanislav Volgushev, Piotr Zwiernik
The practical success of deep learning has led to the discovery of several surprising phenomena. One of these phenomena, that has spurred intense theoretical research, is ``benign…