A fine-tuning and inference strategy combining time-stride autoregression, non-causal temporal upsampling, mean prediction regularization, and gradient difference loss improves generative emulation of PDE simulations on The Well.
Advances in Neural Information Processing Systems , editor=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Kastor: An efficient fine-tuning strategy for generative emulation of PDE simulations
A fine-tuning and inference strategy combining time-stride autoregression, non-causal temporal upsampling, mean prediction regularization, and gradient difference loss improves generative emulation of PDE simulations on The Well.