Summary of Moh: Multi-head Attention As Mixture-of-head Attention, by Peng Jin et al.
MoH: Multi-Head Attention as Mixture-of-Head Attention
by Peng Jin, Bo Zhu, Li Yuan, Shuicheng Yan
First submitted to arxiv on: 15 Oct 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 proposes an upgraded version of the Transformer model’s core mechanism, multi-head attention, which improves efficiency while maintaining or surpassing previous accuracy levels. The authors express multi-head attention in a summation form and propose Mixture-of-Head (MoH) architecture, treating attention heads as experts in the Mixture-of-Experts (MoE) mechanism. MoH enables token selection of appropriate attention heads, enhancing inference efficiency without compromising accuracy or increasing parameters. It also replaces standard summation with weighted summation, introducing flexibility to the attention mechanism and unlocking extra performance potential. Extensive experiments on ViT, DiT, and LLMs demonstrate MoH outperforms multi-head attention by using only 50%-90% of the attention heads. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper improves the Transformer model’s core mechanism, making it more efficient without losing accuracy. It does this by treating different “attention heads” as experts that can be chosen to focus on important parts of the data. This makes the model faster and uses fewer resources. The authors tested their new approach with large language models and found that it performed better than the original method. |
Keywords
» Artificial intelligence » Attention » Inference » Mixture of experts » Multi head attention » Token » Transformer » Vit
