arxivcs.LGcs.AI2026-07-31
HERO: History-Enriched Rollout Training for Long-Horizon Autoregressive Neural Operators
Jiaquan Zhang, Shuxu Chen, Haifan Meng, Yi Lu, Zhihan Lyu, Fan Mo, et al.
Neural operators provide fast surrogates for time-dependent partial differential equations (PDEs) by applying a learned evolution operator recursively to its own predictions, but this autoregressive rollout feeds every prediction error back as input, so local errors accumulate. E…