Summary of Ftmixer: Frequency and Time Domain Representations Fusion For Time Series Modeling, by Zhengnan Li et al.
FTMixer: Frequency and Time Domain Representations Fusion for Time Series Modeling
by Zhengnan Li, Yunxiao Qin, Xilong Cheng, Yuting Tan
First submitted to arxiv on: 24 May 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 This paper proposes a novel architecture, Frequency and Time Domain Mixer (FTMixer), which leverages both the time domain and frequency domain to capture local and global dependencies in time series data. The FTMixer combines the strengths of two key modules: Frequency Channel Convolution (FCC) for capturing global inter-series dependencies and Windowing Frequency Convolution (WFC) for capturing local dependencies. By employing a channel-independent scheme, the WFC module can effectively mix time domain and frequency domain patches to better capture local dependencies. The paper also introduces a novel approach to using Discrete Cosine Transformation (DCT) with real numbers instead of complex numbers in the frequency domain, enabling direct utilization of modern deep learning operators. Experimental results on seven real-world datasets demonstrate the superiority of FTMixer in terms of both forecasting performance and computational efficiency. |
Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper creates a new way to analyze time series data by combining two different methods: one that looks at patterns over time (time domain) and another that looks at patterns across different frequencies (frequency domain). They call this new method FTMixer. The key parts of FTMixer are called FCC and WFC, which help capture global and local dependencies in the data. By mixing these two approaches together, they can create more accurate forecasts and do it more efficiently than before. This new approach uses a special kind of transformation called DCT to make it work with modern deep learning tools. |
Keywords
» Artificial intelligence » Deep learning » Time series