Reproduce “FiLM: Frequency improved Legendre Memory Model for Long-term Time Series Forecasting” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tian Zhou, Ziqing Ma, Xue wang, Qingsong Wen, Liang Sun, Tao Yao, Wotao Yin, Rong Jin
The FiLM model, developed by researchers at Alibaba Group, addresses long-term time series forecasting by integrating Legendre polynomial projections for historical data compression with Fourier analysis for noise removal. This approach yielded an average 20.3% relative MSE reduction over state-of-the-art methods on multivariate benchmarks, while also demonstrating increased computational efficiency and robustness to noise.
Key insights
- Legendre polynomials provide an effective and theoretically grounded method for compressing long historical sequences into a fixed-size representation, accurately preserving essential patterns without excessive noise accumulation over time.
- Fourier analysis is a robust mechanism for distinguishing meaningful signals from noise in time series by leveraging spectral bias, thereby enhancing prediction stability and reliability.
- A synergistic combination of orthogonal polynomial projections for historical data compression and frequency-domain filtering for noise reduction offers a principled and efficient framework for improved long-term time series forecasting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.