Frequency Principle: Fourier Analysis Sheds Light on Deep Neural Networks
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2019
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916254787305472 |
|---|---|
| author | Xu, Zhi-Qin John Zhang, Yaoyu Luo, Tao Xiao, Yanyang Ma, Zheng |
| author_facet | Xu, Zhi-Qin John Zhang, Yaoyu Luo, Tao Xiao, Yanyang Ma, Zheng |
| contents | We study the training process of Deep Neural Networks (DNNs) from the Fourier analysis perspective. We demonstrate a very universal Frequency Principle (F-Principle) -- DNNs often fit target functions from low to high frequencies -- on high-dimensional benchmark datasets such as MNIST/CIFAR10 and deep neural networks such as VGG16. This F-Principle of DNNs is opposite to the behavior of most conventional iterative numerical schemes (e.g., Jacobi method), which exhibit faster convergence for higher frequencies for various scientific computing problems. With a simple theory, we illustrate that this F-Principle results from the regularity of the commonly used activation functions. The F-Principle implies an implicit bias that DNNs tend to fit training data by a low-frequency function. This understanding provides an explanation of good generalization of DNNs on most real datasets and bad generalization of DNNs on parity function or randomized dataset. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_1901_06523 |
| institution | arXiv |
| publishDate | 2019 |
| record_format | arxiv |
| spellingShingle | Frequency Principle: Fourier Analysis Sheds Light on Deep Neural Networks Xu, Zhi-Qin John Zhang, Yaoyu Luo, Tao Xiao, Yanyang Ma, Zheng Machine Learning 68Q32, 68T01 I.2.6 We study the training process of Deep Neural Networks (DNNs) from the Fourier analysis perspective. We demonstrate a very universal Frequency Principle (F-Principle) -- DNNs often fit target functions from low to high frequencies -- on high-dimensional benchmark datasets such as MNIST/CIFAR10 and deep neural networks such as VGG16. This F-Principle of DNNs is opposite to the behavior of most conventional iterative numerical schemes (e.g., Jacobi method), which exhibit faster convergence for higher frequencies for various scientific computing problems. With a simple theory, we illustrate that this F-Principle results from the regularity of the commonly used activation functions. The F-Principle implies an implicit bias that DNNs tend to fit training data by a low-frequency function. This understanding provides an explanation of good generalization of DNNs on most real datasets and bad generalization of DNNs on parity function or randomized dataset. |
| title | Frequency Principle: Fourier Analysis Sheds Light on Deep Neural Networks |
| topic | Machine Learning 68Q32, 68T01 I.2.6 |
| url | https://arxiv.org/abs/1901.06523 |