Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Old Optimizer, New Norm: An Anthology

About

Deep learning optimizers are often motivated through a mix of convex and approximate second-order theory. We select three such methods -- Adam, Shampoo and Prodigy -- and argue that each method can instead be understood as a squarely first-order method without convexity assumptions. In fact, after switching off exponential moving averages, each method is equivalent to steepest descent under a particular norm. By generalizing this observation, we chart a new design space for training algorithms. Different operator norms should be assigned to different tensors based on the role that the tensor plays within the network. For example, while linear and embedding layers may have the same weight space of $\mathbb{R}^{m\times n}$, these layers play different roles and should be assigned different norms. We hope that this idea of carefully metrizing the neural architecture might lead to more stable, scalable and indeed faster training.

Jeremy Bernstein, Laker Newhouse• 2024

Related benchmarks

TaskDatasetResultRank
Image ClassificationCIFAR-10 (val)--
377
Matrix completionMovieLens 1M (test)
RMSE0.8693
37
ClassificationYouTube Faces
Mean Accuracy26.4
32
SPD covariance prototype classificationCIFAR-100 coarse labels (test)
Mean Test Accuracy40.7
6
Language ModelingFineWeb-Edu 10B (val)
Validation CE3.27
5
Showing 5 of 5 rows

Other info

Follow for update