Summary of Rethinking Softmax: Self-attention with Polynomial Activations, by Hemanth Saratchandran et al.
Rethinking Softmax: Self-Attention with Polynomial Activations
by Hemanth Saratchandran, Jianqiao Zheng, Yiping Ji, Wenbo Zhang, Simon Lucey
First submitted to arxiv on: 24 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (stat.ML)
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 research paper challenges a widely held assumption about the effectiveness of softmax attention in transformers. Instead of generating a probability distribution for attention allocation, its success is due to its ability to implicitly regularize the Frobenius norm of the attention matrix during training. The authors theoretically show this and explore alternative activations that can achieve this effect, demonstrating that certain polynomial activations are suitable for attention-based architectures. Empirical results indicate these activations perform comparably or better than softmax across various computer vision and language tasks, suggesting new possibilities for attention mechanisms beyond softmax. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper shows that a popular way to use transformers is actually working because of an unexpected reason. Instead of helping with attention, the “softmax” technique is helping by making sure the attention weights are not too extreme. The authors come up with new ways to do this and test them on different tasks. They find that these new methods work just as well or even better than the old way, which means we can try new things in the future. |
Keywords
» Artificial intelligence » Attention » Probability » Softmax
