Reproduce “TARFVAE: Efficient One-Step Generative Time Series Forecasting via TARFLOW based VAE” 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
Jiawen Wei, Lan Jiang, Pengbo Wei, Ziwen Ye, Teng Song, Chen Chen, Guangrui Ma
TARFVAE uses a training-time autoregressive flow to refine a variational autoencoder’s latent representation for long-horizon probabilistic forecasting. At inference, it samples from a history-conditioned prior and decodes the full future in one pass, avoiding the flow’s sequential inverse. On ETTh1 at horizon 720, one-sample inference takes 3.0 ms versus mr-Diff’s 21.6 ms, a benchmark-specific result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.