Summary of Frednormer: Frequency Domain Normalization For Non-stationary Time Series Forecasting, by Xihao Piao et al.
FredNormer: Frequency Domain Normalization for Non-stationary Time Series Forecasting
by Xihao Piao, Zheng Chen, Yushun Dong, Yasuko Matsubara, Yasushi Sakurai
First submitted to arxiv on: 2 Oct 2024
Categories
- Main: Machine Learning (stat.ML)
- Secondary: 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 presents a novel approach to improving time series forecasting by analyzing frequency components in datasets. Recent normalization-based methods have shown success in tackling distribution shifts, but they operate in the time domain and may fail to capture dynamic patterns in the frequency domain. The authors prove that current normalization methods uniformly scale non-zero frequencies, leading to suboptimal results. They propose FredNormer, a plug-and-play module that observes datasets from a frequency perspective, adaptively up-weights key frequency components, and introduces sample-specific variations. Experiments show that FredNormer improves the averaged MSE of backbone forecasting models by 33.3% and 55.3% on the ETTm2 dataset. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps improve time series forecasting by looking at datasets in a new way. Normalization methods have been successful, but they don’t always work well because they only look at how things change over time. The authors show that current methods can miss important patterns that are more visible when you look at data in the frequency domain. They create a new module called FredNormer that does this and helps improve forecasts. |
Keywords
» Artificial intelligence » Mse » Time series
