Reproduce “Cross-View Time-Frequency Diffusion for Extreme-Aware Time-Series Generation” 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
Lin Jiang, Rongchao Xu, Guang Wang
CrossTF generates time series with improved fidelity to rare events by adjusting between two denoiser estimates of a sequence, one based on time structure and one on frequency structure. It tunes six values during late denoising while keeping the denoisers frozen. Across seven datasets, extreme fidelity improves by an average 26.1% versus strongest baselines, while overall performance declines on three datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.