Reproduce “Time series generation with spectrally aligned latent flow matching” 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
Camilo Carvajal Reyes, Felipe Tobar
The paper improves latent flow matching for unconditional time-series generation by fine-tuning an autoencoder with transform-consistency losses before training the flow. Fourier, wavelet, or signature features receive detail-sensitive weighting so compressed codes preserve signal structure. On Weather, wavelet alignment achieves a discriminative score of 0.282±0.097 versus 0.379±0.063 for the base model, but gains vary across datasets, transforms, and metrics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.