Reproduce “Fast Temporal Wavelet Graph Neural Networks” 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
Duc Thien Nguyen, Manh Duc Tuan Nguyen, Truong Son Hy, Risi Kondor
FTWGNN uses MMF-based graph wavelet convolution in a recurrent encoder-decoder to forecast signals over connected sensors. Its sparse basis represents graph relationships at multiple scales, replacing DCRNN’s diffusion convolution while retaining recurrent forecasting. On PEMS-BAY, its 60-minute mean absolute error is 1.79 versus DCRNN’s 2.07. On METR-LA, FTWGNN trains in 136 seconds per epoch versus DCRNN’s 1,800, a reported 13.23-fold speedup; this comparison does not isolate convolution effects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.