Reproduce “TimeWorld: Towards Action-Conditioned Multimodal World Models for Time Series Forecasting” 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
Hang Ni, Weijia Zhang, Hao Liu
TimeWorld forecasts observed outcomes under recorded actions by adding action-specific updates to future representations of a pretrained time-series forecasting model. It also uses contrastive training to distinguish action types and intensities. Across four backbones, it reports lower aggregate normalized mean absolute error than history-only forecasting and text-prefix adaptation. This evaluates recorded-action forecasts, not verified outcomes for unobserved interventions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.