Summary of Dominant Shuffle: a Simple Yet Powerful Data Augmentation For Time-series Prediction, by Kai Zhao et al.
Dominant Shuffle: A Simple Yet Powerful Data Augmentation for Time-series Prediction
by Kai Zhao, Zuojie He, Alex Hung, Dan Zeng
First submitted to arxiv on: 26 May 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 explores frequency-domain data augmentation (DA) for time series prediction. Researchers have shown that DA can be effective, but existing methods disturb the original data with full-spectrum noises, leading to a domain gap between augmented and original data. To address this, the authors propose two modifications to improve frequency-domain DA. First, they limit perturbations to dominant frequencies, which represent the main periodicity and trends of the signal. Second, they suggest shuffling dominant frequency components instead of designed random perturbations. This simple yet powerful method, called dominant shuffle, can be implemented with just a few lines of code. The authors conduct extensive experiments with eight datasets and six time series models, demonstrating that dominant shuffle consistently improves baseline performance and outperforms other DA methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Time series prediction is important for many applications! Researchers have been using something called frequency-domain data augmentation to make predictions better. But they’ve found that this method can be tricky because it adds a lot of noise to the original data. To fix this, scientists are proposing two new ways to do frequency-domain DA. First, they’re only adding noise to the most important parts of the signal. Second, they’re just mixing up those important parts instead of adding random noise. This new method is easy to use and can make predictions better. It’s been tested on many different datasets and it works! |
Keywords
» Artificial intelligence » Data augmentation » Time series
