Summary of Hybrid Of Diffstride and Spectral Pooling in Convolutional Neural Networks, by Sulthan Rafif et al.
Hybrid of DiffStride and Spectral Pooling in Convolutional Neural Networks
by Sulthan Rafif, Mochamad Arfan Ravy Wahyu Pratama, Mohammad Faris Azhar, Ahmad Mustafidul Ibad, Lailil Muflikhah, Novanto Yudistira
First submitted to arxiv on: 17 Jan 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 The proposed CNN model combines the learnable stride technique with spectral pooling to improve image classification accuracy. By allowing the network to learn its own stride value, the DiffStride method can capture important information contained in images that might be missed by fixed strides. The downsampling process is a crucial step in convolutional neural networks (CNNs), but it often results in severe quantization and a constraining lower bound on preserved information. Spectral pooling reduces this constraint by cutting off the representation in the frequency domain, allowing for more effective Downsampling Learnable Stride Technique performed through backpropagation combined with spectral pooling. The hybrid method, which combines spectral pooling and DiffStride, outperforms the baseline method by 0.0094 in accuracy on ResNet-18. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new way to make computer images better is being explored. Some methods for downsampling (making smaller) are not very good because they throw away important information. The researchers created a new combination of techniques that helps keep more of this information. They tested it and found that it was 0.0094% better than the old method at recognizing objects in pictures. |
Keywords
» Artificial intelligence » Backpropagation » Cnn » Image classification » Quantization » Resnet
