Summary of A Greedy Hierarchical Approach to Whole-network Filter-pruning in Cnns, by Kiran Purohit et al.
A Greedy Hierarchical Approach to Whole-Network Filter-Pruning in CNNs
by Kiran Purohit, Anurag Reddy Parvathgari, Sourangshu Bhattacharya
First submitted to arxiv on: 22 Aug 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: 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 A deep learning paper proposes a hierarchical approach to prune redundant filters in convolutional neural networks (CNNs), enabling efficient models for resource-constrained devices. The method, which combines two algorithms, filter-pruning and layer-selection, leverages the classification loss as the final criterion. Filter-pruning employs linear approximation of filter weights, while layer-selection greedily selects the best-pruned layer using global pruning criteria. The proposed approach outperforms state-of-the-art pruning methods on various CNN architectures, including ResNet18, ResNet32, ResNet56, VGG16, and ResNext101. The method reduces RAM requirements for ResNext101 from 7.6 GB to 1.5 GB and achieves a 94% reduction in floating-point operations (FLOPS) without compromising accuracy on CIFAR-10. |
Low | GrooveSquid.com (original content) | Low Difficulty Summary A new approach to make computer models smaller and more efficient is developed. This helps devices with limited resources run complex tasks like image recognition faster and use less power. The method combines two steps: one that prunes filters in each layer of the model, and another that selects the best layers to prune. The process uses a measure called classification loss to decide which filters or layers to remove. The result is a smaller model that still performs well on tasks like recognizing objects in images. |
Keywords
» Artificial intelligence » Classification » Cnn » Deep learning » Pruning