Summary of Convolutional Neural Network Compression Via Dynamic Parameter Rank Pruning, by Manish Sharma et al.
Convolutional Neural Network Compression via Dynamic Parameter Rank Pruning
by Manish Sharma, Jamison Heard, Eli Saber, Panos P. Markopoulos
First submitted to arxiv on: 15 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 This paper addresses a critical issue in Convolutional Neural Networks (CNNs): over-parameterization, which leads to reduced performance and limited deployment on edge devices. The authors propose an efficient training method for CNN compression via dynamic parameter rank pruning. Their approach combines low-rank matrix approximation and novel regularization techniques, leveraging Singular Value Decomposition (SVD) to model low-rank convolutional filters and dense weight matrices. By training the SVD factors with back-propagation in an end-to-end way, they achieve substantial storage savings while maintaining or even enhancing classification performance on popular CNNs like ResNet-18, ResNet-20, and ResNet-32. The proposed method demonstrates its applicability in computer vision tasks using datasets such as CIFAR-10, CIFAR-100, and ImageNet (2012). |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps solve a problem with special kinds of AI models called Convolutional Neural Networks (CNNs). These models are very good at recognizing patterns, but they can be too big and use too much power to work on small devices like smartphones. The authors developed a new way to make these models smaller without losing their ability to recognize things correctly. They used a special math technique called Singular Value Decomposition (SVD) to shrink the models while keeping them accurate. This approach works well with popular image recognition tasks and can be used in many real-world applications. |
Keywords
» Artificial intelligence » Classification » Cnn » Pruning » Regularization » Resnet
