Peter Bartlett

FutureFill: Fast Generation from Convolutional Sequence Models

A Mathematical Framework for the Hierarchical Analysis of Neural Networks

Improved Scaling Laws in Linear Regression via Data Reuse

Benefits of Early Stopping in Gradient Descent for Overparameterized Logistic Regression

Large Stepsizes Accelerate Gradient Descent for Regularized Logistic Regression

Implicit Bias of Gradient Descent for Non-Homogeneous Deep Networks

Gradient Descent Converges Arbitrarily Fast for Logistic Regression via Large and Adaptive Stepsizes

Provable Length Generalization in Sequence Prediction via Spectral Filtering

FutureFill: Fast Generation from Convolutional Sequence Models