Tag
This paper benchmarks the rollout reliability of seven PDE foundation models across seven 2D time-dependent PDE datasets, showing that while fine-tuned foundation models often beat scratch-trained baselines at short horizons, full-horizon rollouts suffer substantial error amplification (median 2.04x), arguing for horizon-resolved reporting in future benchmarks.
This paper analyzes long-horizon rollout error in Graph World Models (GWMs), proposing a unified framework with dynamic edges and introducing Error-Aware GWM that uses spectral regularization, rollout consistency, and critical-node weighting to prevent divergence.