Reproduce “PAMNet: Cycle-aware Phase-Amplitude Modulation Network for Multivariate 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

Yingbo Zhou, Yutong Ye, Zhiwei Ling, Shuhao Li, Rui Qian, Jian Xiong, Li Sun, Dejing Dou

PAMNet, a Cycle-aware Phase-Amplitude Modulation Network, explicitly decomposes periodic patterns in multivariate time series into distinct phase and amplitude components, modeling their non-stationary coupling. This approach achieved top-1 performance in 20 out of 24 multivariate forecasting metrics across twelve real-world datasets, offering both enhanced accuracy and computational efficiency.

Key insights

  • Explicitly decoupling and modulating phase and amplitude components provides a more robust and nuanced mechanism for handling non-stationary periodic patterns in multivariate time series.
  • Learnable, cycle-aware carrier embeddings can adaptively capture precise temporal positioning ('phase') and intensity modulation ('amplitude') of events within a cycle, enhancing model flexibility.
  • The lightweight dual-path modulation design offers computational efficiency by performing explicit linear-complexity phase-amplitude disentanglement, avoiding the quadratic costs of global attention.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.