ICML 2021spotlight76 citations

Tensor Programs IIb: Architectural Universality Of Neural Tangent Kernel Training Dynamics

Greg Yang, Etai Littwin

Abstract

Yang (2020) recently showed that the Neural Tangent Kernel (NTK) at initialization has an infinite-width limit for a large class of architectures including modern staples such as ResNet and Transformers. However, their analysis does not apply to training. Here, we show the same neural networks (in the so-called NTK parametrization) during training follow a kernel gradient descent dynamics in function space, where the kernel is the infinite-width NTK. This completes the proof of the architectural universality of NTK behavior. To achieve this result, we apply the Tensor Programs technique: Write the entire SGD dynamics inside a Tensor Program and analyze it via the Master Theorem. To facilitate this proof, we develop a graphical notation for Tensor Programs, which we believe is also an important contribution toward the pedagogy and exposition of the Tensor Programs technique.

BibTeX
@InProceedings{pmlr-v139-yang21f,
  title = 	 {Tensor Programs IIb: Architectural Universality Of Neural Tangent Kernel Training Dynamics},
  author =       {Yang, Greg and Littwin, Etai},
  booktitle = 	 {Proceedings of the 38th International Conference on Machine Learning},
  pages = 	 {11762--11772},
  year = 	 {2021},
  editor = 	 {Meila, Marina and Zhang, Tong},
  volume = 	 {139},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {18--24 Jul},
  publisher =    {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v139/yang21f/yang21f.pdf},
  url = 	 {https://proceedings.mlr.press/v139/yang21f.html},
  abstract = 	 {Yang (2020) recently showed that the Neural Tangent Kernel (NTK) at initialization has an infinite-width limit for a large class of architectures including modern staples such as ResNet and Transformers. However, their analysis does not apply to training. Here, we show the same neural networks (in the so-called NTK parametrization) during training follow a kernel gradient descent dynamics in function space, where the kernel is the infinite-width NTK. This completes the proof of the architectural universality of NTK behavior. To achieve this result, we apply the Tensor Programs technique: Write the entire SGD dynamics inside a Tensor Program and analyze it via the Master Theorem. To facilitate this proof, we develop a graphical notation for Tensor Programs, which we believe is also an important contribution toward the pedagogy and exposition of the Tensor Programs technique.}
}