Summary of Token Statistics Transformer: Linear-time Attention Via Variational Rate Reduction, by Ziyang Wu et al.
Token Statistics Transformer: Linear-Time Attention via Variational Rate Reduction
by Ziyang Wu, Tianjiao Ding, Yifu Lu, Druv Pai, Jingyuan Zhang, Weida Wang, Yaodong Yu, Yi Ma, Benjamin D. Haeffele
First submitted to arxiv on: 23 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 proposed novel Transformer attention operator, Token Statistics Self-Attention (TSSA), boasts linear computational and memory complexity, a significant departure from the typical quadratic scaling of standard self-attention. By extending prior work on white-box architecture design and deriving a variational form of the maximal coding rate reduction objective, the authors derive TSSA, which achieves competitive performance with conventional Transformers while being more computationally efficient and interpretable. The Token Statistics Transformer (ToST), which swaps TSSA for standard self-attention, demonstrates state-of-the-art results on vision, language, and long sequence tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper introduces a new attention operator called Token Statistics Self-Attention (TSSA) that can process sequences more efficiently than traditional transformer architectures. The authors show that by using this new operator, they can achieve similar results to the original transformers but with much less computational effort required. This could be important for large-scale applications where processing speed is crucial. |
Keywords
» Artificial intelligence » Attention » Self attention » Token » Transformer
