Summary of Unveiling and Controlling Anomalous Attention Distribution in Transformers, by Ruiqing Yan et al.
Unveiling and Controlling Anomalous Attention Distribution in Transformers
by Ruiqing Yan, Xingbo Du, Haoyu Deng, Linghan Zheng, Qiuzhuang Sun, Jifang Hu, Yuhang Shao, Penghao Jiang, Jinrong Jiang, Lian Zhao
First submitted to arxiv on: 26 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- 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 The Transformer architecture’s Attention mechanism has been observed to exhibit an anomalous behavior, where a significant amount of attention is concentrated on the first element. This phenomenon affects large models based on Transformers, making it crucial to understand its cause for developing techniques like Key-Value Cache compression and infinite extrapolation. Researchers have found that the high attention on the first element can be attributed to waiver phenomena, which involve reducing internal values in certain sequence elements to absorb excess attention without affecting their contribution to information. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This phenomenon affects large models based on Transformers, making it crucial to understand its cause for developing techniques like Key-Value Cache compression and infinite extrapolation. Researchers have found that the high attention on the first element can be attributed to waiver phenomena, which involve reducing internal values in certain sequence elements to absorb excess attention without affecting their contribution to information. |
Keywords
* Artificial intelligence * Attention * Transformer
