Summary of Learning Convolutional Neural Networks in the Frequency Domain, by Hengyue Pan and Yixin Chen and Xin Niu and Wenbo Zhou and Dongsheng Li
Learning Convolutional Neural Networks in the Frequency Domain
by Hengyue Pan, Yixin Chen, Xin Niu, Wenbo Zhou, Dongsheng Li
First submitted to arxiv on: 14 Apr 2022
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes the CEMNet, a novel convolutional neural network (CNN) that can be trained in the frequency domain. By leveraging the Cross-Correlation Theorem, the image convolution operation is replaced with element-wise multiplication, significantly reducing computation complexity. To alleviate over-fitting, the Weight Fixation mechanism is introduced, and counterparts for Batch Normalization, Leaky ReLU, and Dropout are designed to optimize performance in the frequency domain. A two-branches network structure accommodates complex inputs from Discrete Fourier Transform. Experimental results on MNIST and CIFAR-10 databases demonstrate CEMNet’s effectiveness. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper creates a new way for computers to recognize images using a special kind of neural network called the CEMNet. The old way used something called image convolution, which is very good at recognizing patterns in pictures, but it takes a lot of computer power and can be slow. The researchers found a mathematical trick that lets them use a much faster operation instead. They also developed new ways to prevent the computer from getting too good at recognizing patterns (which can make it bad at generalizing), and made sure their system could handle complicated images. When they tested their CEMNet on pictures of handwritten numbers and objects, it did very well. |
Keywords
* Artificial intelligence * Batch normalization * Cnn * Dropout * Neural network * Relu
