Summary of Long Input Sequence Network For Long Time Series Forecasting, by Chao Ma et al.
Long Input Sequence Network for Long Time Series Forecasting
by Chao Ma, Yikai Hou, Xiang Li, Yinggang Sun, Haining Yu
First submitted to arxiv on: 18 Jul 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 This paper tackles a common challenge in deep learning-based long-term time-series forecasting: fixed-length inputs leading to overfitting and decreased accuracy. The authors identify the root cause as the interaction between multi-scale patterns in the time series and the fixed focus scale of current models. They propose a novel approach, decoupling these patterns by modeling each with its corresponding period length as token size. This is achieved through a Series-Decomposition Module (MPSD) and Multi-Token Pattern Recognition Neural Network (MTPR). These innovations enable models to handle inputs up to 10 times longer than before, resulting in significant performance improvements (38% maximum precision increase), while maintaining low complexity (0.22 times the cost) and high interpretability. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research solves a problem with deep learning for long-term forecasting. Right now, most methods have limitations when dealing with very long sequences of data. The authors figured out that this is because current models focus too much on small details and don’t take into account the bigger patterns in the data. They developed new techniques to handle longer inputs by breaking down the big patterns into smaller pieces, each related to a specific time period. This allows for better performance (38% improvement) while keeping things simple and easy to understand. |
Keywords
» Artificial intelligence » Deep learning » Neural network » Overfitting » Pattern recognition » Precision » Time series » Token
