Summary of Exploring the Role Of Token in Transformer-based Time Series Forecasting, by Jianqi Zhang et al.
Exploring the Role of Token in Transformer-based Time Series Forecasting
by Jianqi Zhang, Jingyao Wang, Chuxiong Sun, Xingchen Shen, Fanjiang Xu, Changwen Zheng, Wenwen Qiang
First submitted to arxiv on: 16 Apr 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- 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 paper explores the role of temporal or variable tokens in transformer-based time series forecasting (TSF) models. While most studies focus on optimizing model structure, this research highlights the importance of token selection for effective predictions. Through theoretical analyses and experiments, the authors find that gradients mainly depend on “positive tokens” that contribute to predicted series. They also identify factors that help models select these positive tokens, such as positional encoding (PE) and its interaction with network depth. Inspired by these findings, the authors propose temporal positional encoding (T-PE) and variable positional encoding (V-PE), which are used in a Transformer-based dual-branch framework called T2B-PE. Experimental results demonstrate the superiority of T2B-PE in terms of robustness and effectiveness. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper is about how to make better predictions when forecasting things that change over time, like weather or stock prices. It’s all about understanding which pieces of information are important for making good predictions. The researchers found that some types of information help models make better choices, while other types get weaker as the model gets more complicated. They used this knowledge to create new ways of using information and tested them on different forecasting tasks. The results show that their new approach is really good at making accurate predictions. |
Keywords
» Artificial intelligence » Positional encoding » Time series » Token » Transformer
