Transition to Linearity of General Neural Networks with Directed Acyclic Graph Architecture
Abstract
In this paper we show that feedforward neural networks corresponding to arbitrary directed acyclic graphs undergo transition to linearity as their ``width'' approaches infinity. The width of these general networks is characterized by the minimum in-degree of their neurons, except for the input and first layers. Our results identify the mathematical structure underlying transition to linearity and generalize a number of recent works aimed at characterizing transition to linearity or constancy of the Neural Tangent Kernel for standard architectures.
Cite
Text
Zhu et al. "Transition to Linearity of General Neural Networks with Directed Acyclic Graph Architecture." Neural Information Processing Systems, 2022.Markdown
[Zhu et al. "Transition to Linearity of General Neural Networks with Directed Acyclic Graph Architecture." Neural Information Processing Systems, 2022.](https://mlanthology.org/neurips/2022/zhu2022neurips-transition/)BibTeX
@inproceedings{zhu2022neurips-transition,
title = {{Transition to Linearity of General Neural Networks with Directed Acyclic Graph Architecture}},
author = {Zhu, Libin and Liu, Chaoyue and Belkin, Misha},
booktitle = {Neural Information Processing Systems},
year = {2022},
url = {https://mlanthology.org/neurips/2022/zhu2022neurips-transition/}
}