Recurrently training neural operators on their own predictions reduces long-term forecast error and error growth compared with teacher forcing, though the theoretical linear-growth proof depends on an unproven assumption.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Recurrent Neural Operators: Stable Long-Term PDE Prediction
Recurrently training neural operators on their own predictions reduces long-term forecast error and error growth compared with teacher forcing, though the theoretical linear-growth proof depends on an unproven assumption.