Reproduce “Seq-Flow: Efficient Probabilistic Forecasting with Self-Rollout Error Control” 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
Yinan Huang, Shitij Govil, Bo Dai, Pan Li
Seq-Flow updates probabilistic forecasts by transporting samples from a previous forecast toward a distribution conditioned on newly observed history, rather than restarting each update from noise. It trains on model-generated forecast rollouts and partially renoises their sources. On Beam Spill, at 3 NFE, it achieves CRPS 0.015 versus 0.043 for full-trajectory diffusion, but this measures forecast quality, not runtime.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.