Saved in:
Bibliographic Details
Main Authors: Loeschcke, Sebastian, Pitt, David, George, Robert Joseph, Zhao, Jiawei, Luo, Cheng, Tian, Yuandong, Kossaifi, Jean, Anandkumar, Anima
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2501.02379
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915314791350272
author Loeschcke, Sebastian
Pitt, David
George, Robert Joseph
Zhao, Jiawei
Luo, Cheng
Tian, Yuandong
Kossaifi, Jean
Anandkumar, Anima
author_facet Loeschcke, Sebastian
Pitt, David
George, Robert Joseph
Zhao, Jiawei
Luo, Cheng
Tian, Yuandong
Kossaifi, Jean
Anandkumar, Anima
contents Scientific problems require resolving multi-scale phenomena across different resolutions and learning solution operators in infinite-dimensional function spaces. Neural operators provide a powerful framework for this, using tensor-parameterized layers to capture complex, multi-dimensional relationships. However, scaling neural operators to high-resolution problems leads to significant computational demands, making the training of industrial-scale models prohibitive. In this work, we introduce \textbf{TensorGRaD}, a novel method that directly addresses the memory challenges associated with optimizing large tensor-structured weights. Our approach, based on a \texit{robust tensor decomposition}, factorizes gradients as the sum of a low-rank tensor and a sparse one to efficiently capture information within optimizer states, including outliers. Additionally, we provide a recipe for mixed precision training of TensorGRaD, achieving further memory savings without sacrificing accuracy. We showcase the effectiveness of TensorGRaD on Fourier Neural Operators, a class of models crucial for solving partial differential equations (PDE). We provide theoretical guarantees for TensorGRaD, demonstrating its fundamental advantage over matrix-based gradient compression methods. We empirically demonstrate large improvements across various PDE tasks, including the challenging turbulent Navier-Stokes case at a Reynolds number of $10^5$. TensorGRaD reduces total memory usage by over $50\%$ while maintaining and sometimes even improving accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2501_02379
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle TensorGRaD: Tensor Gradient Robust Decomposition for Memory-Efficient Neural Operator Training
Loeschcke, Sebastian
Pitt, David
George, Robert Joseph
Zhao, Jiawei
Luo, Cheng
Tian, Yuandong
Kossaifi, Jean
Anandkumar, Anima
Machine Learning
Scientific problems require resolving multi-scale phenomena across different resolutions and learning solution operators in infinite-dimensional function spaces. Neural operators provide a powerful framework for this, using tensor-parameterized layers to capture complex, multi-dimensional relationships. However, scaling neural operators to high-resolution problems leads to significant computational demands, making the training of industrial-scale models prohibitive. In this work, we introduce \textbf{TensorGRaD}, a novel method that directly addresses the memory challenges associated with optimizing large tensor-structured weights. Our approach, based on a \texit{robust tensor decomposition}, factorizes gradients as the sum of a low-rank tensor and a sparse one to efficiently capture information within optimizer states, including outliers. Additionally, we provide a recipe for mixed precision training of TensorGRaD, achieving further memory savings without sacrificing accuracy. We showcase the effectiveness of TensorGRaD on Fourier Neural Operators, a class of models crucial for solving partial differential equations (PDE). We provide theoretical guarantees for TensorGRaD, demonstrating its fundamental advantage over matrix-based gradient compression methods. We empirically demonstrate large improvements across various PDE tasks, including the challenging turbulent Navier-Stokes case at a Reynolds number of $10^5$. TensorGRaD reduces total memory usage by over $50\%$ while maintaining and sometimes even improving accuracy.
title TensorGRaD: Tensor Gradient Robust Decomposition for Memory-Efficient Neural Operator Training
topic Machine Learning
url https://arxiv.org/abs/2501.02379