https://github.com/brain-research/conv-sv

The Singular Values of Convolutional Layers
https://github.com/brain-research/conv-sv

Last synced: 3 months ago
JSON representation

The Singular Values of Convolutional Layers

• Host: GitHub
• URL: https://github.com/brain-research/conv-sv
• Owner: brain-research
• License: apache-2.0
• Archived: true
• Created: 2018-09-25T22:41:00.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2018-09-25T23:40:35.000Z (about 6 years ago)
• Last Synced: 2024-07-04T01:02:14.084Z (4 months ago)
• Language: Python
• Homepage: https://arxiv.org/pdf/1805.10408
• Size: 12.7 KB
• Stars: 71
• Watchers: 6
• Forks: 7
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • Contributing: CONTRIBUTING.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# The Singular Values of Convolutional Layers  #

**Link to paper:[https://arxiv.org/pdf/1805.10408](https://arxiv.org/pdf/1805.10408 "The Singular Values of Convolutional Layers")**

Overview 

======

 

We characterize the singular values of the linear transformation associated with a convolution applied to a two-dimensional feature map with multiple channels. Our characterization enables efficient computation of the singular values of convolutional layers used in popular deep neural network architectures. It also leads to an algorithm for projecting a convolutional layer onto the set of layers obeying a bound on the operator norm of the layer.

Here, we provide the code for

* Our new method for calculating the singular values for 2D multi-channel conovlutional layers.	

* Time test comparing different methods for calculating the singular values.	

* Sketching the singular values for any trained model.	

* projecting a convolutional layer onto the set of layers obeying a bound on the operator norm of the layer.

	

### How do I get set up? ###

Requirements

+ Tensorflow

+ Numpy

## Citation ##

If you use this code, please cite our paper:

```

@article{

  sedghi2018singular,

  title={The Singular Values of Convolutional Layers},

  author={Sedghi, Hanie and Gupta, Vineet and Long, Philip M},

  journal={arXiv preprint arXiv:1805.10408},

  year={2018}

}

```

Authors 

======

* Hanie Sedghi ([[email protected]]([email protected] "mailto:[email protected]"))

* Vineet Gupta ([[email protected]]([email protected] "mailto:[email protected]"))

* Phil Long ([[email protected]]([email protected] "mailto:[email protected]"))

This is not an officially supported Google Product.