Reproduce “Timestral” 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
Adrien Petralia, Alexandre Gavaudan, Artjom Joosen, Etienne Goffinet, Richard Kurle, Yassine El Ouahidi, Yassir Bendou, Abdelaziz Bounhar and 348 more
Timestral is a time-series forecasting model that learns across series and forecasts new ones without retraining. It keeps observed history available while adding separate prediction positions, and assigns covariates roles so known-future inputs can be used without exposing future targets. Timestral-209M ranks first among compared models on six metrics across three benchmarks; this does not isolate DHF’s contribution.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.