A major challenge in training large-scale machine learning models is configuring the training process to maximize model performance, i.e., finding the best training setup from a vast design space. In this work, we unlock a gradient-based approach to this problem. We first introduce an algorithm for efficiently calculating metagradients -- gradients through model training -- at scale. We then introduce a "smooth model training" framework that enables effective optimization using metagradients. With metagradient descent (MGD), we greatly improve on existing dataset selection methods, outperform accuracy-degrading data poisoning attacks by an order of magnitude, and automatically find competitive learning rate schedules.
View on arXiv@article{engstrom2025_2503.13751,
title={ Optimizing ML Training with Metagradient Descent },
author={ Logan Engstrom and Andrew Ilyas and Benjamin Chen and Axel Feldmann and William Moses and Aleksander Madry },
journal={arXiv preprint arXiv:2503.13751},
year={ 2025 }
}