Summary of When Training-free Nas Meets Vision Transformer: a Neural Tangent Kernel Perspective, by Qiqi Zhou et al.
When Training-Free NAS Meets Vision Transformer: A Neural Tangent Kernel Perspective
by Qiqi Zhou, Yichen Zhu
First submitted to arxiv on: 15 Mar 2024
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 The paper investigates the Neural Tangent Kernel (NTK) in searching vision transformers without training. Unlike previous observations, which found NTK-based metrics effective in predicting CNNs performance at initialization, this study shows that NTK is inefficient for ViT search spaces. The authors hypothesize that feature learning preferences within ViT contribute to this inefficacy and propose a new method called ViNTK that generalizes standard NTK to the high-frequency domain by integrating Fourier features from inputs. Experiments on image classification and semantic segmentation tasks demonstrate that ViNTK can significantly speed up search costs while maintaining similar performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how Neural Tangent Kernel (NTK) works in searching vision transformers without training. Most people think NTK is good for predicting what a computer will do when it’s first turned on, but this study shows that doesn’t work as well with vision transformers. The authors think that the way vision transformers learn new features makes it hard for NTK to work well. To fix this problem, they created a new method called ViNTK that can handle high-frequency signals in feature learning. They tested ViNTK on two tasks and found that it can search much faster while still getting good results. |
Keywords
» Artificial intelligence » Image classification » Semantic segmentation » Vit
