Summary of Data-independent Module-aware Pruning For Hierarchical Vision Transformers, by Yang He et al.
Data-independent Module-aware Pruning for Hierarchical Vision Transformers
by Yang He, Joey Tianyi Zhou
First submitted to arxiv on: 21 Apr 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 Hierarchical vision transformers (ViTs) have advantages over conventional ViTs. They achieve linear computational complexity by local self-attention and create hierarchical feature maps for dense prediction. However, existing pruning methods ignore these unique properties and use magnitude values as weight importance. This approach has two drawbacks: comparing “local” attention weights at a “global” level can prune important weights, and it doesn’t consider distinct weight distributions essential for extracting features. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Hierarchical vision transformers (ViTs) are new types of computer models that are better than old ones in some ways. They can process images quickly and create detailed maps of what’s inside the image. But when we try to make these models smaller, the current methods don’t work well because they compare small parts with big parts and ignore how each part is different from others. |
Keywords
» Artificial intelligence » Attention » Pruning » Self attention
