AID | 让样本主动“找难点”：RAMS提升偏微分方程科学机器学习的精度与效率
Ranking
Overall
66
Content
70
Popularity
55
Observed public metrics from 1 member.
Merged summary
TL;DR - RAMS is a residual-driven adversarial-gradient sampling method that moves existing training samples toward regions where PDE models have high residuals. It improves PINN and neural-operator accuracy and sample efficiency with little added computation, especially in high-dimensional problems.
- Treats coordinates or input functions as trainable parameters and alternates model-loss minimization with residual-maximizing sample updates.
- Integrates with existing sampling methods and supports PINNs, physics-informed operators, and data-driven operator learning.
- Reduced errors by 40–95% in reported Poisson experiments, while sample-movement overhead averaged under 2% of runtime.
- Maintained error below (10^{-2}) on a 10-dimensional PDE and achieved comparable operator-learning accuracy with substantially fewer samples than cited baselines.
Sources (1)
AID | 让样本主动“找难点”：RAMS提升偏微分方程科学机器学习的精度与效率
Public signals
OpenAlex citations 3
TL;DR - RAMS is a residual-driven adversarial-gradient sampling method that moves existing training samples toward regions where PDE models have high residuals. It improves PINN and neural-operator accuracy and sample efficiency with little added computation, especially in high-dimensional problems.
- Treats coordinates or input functions as trainable parameters and alternates model-loss minimization with residual-maximizing sample updates.
- Integrates with existing sampling methods and supports PINNs, physics-informed operators, and data-driven operator learning.
- Reduced errors by 40–95% in reported Poisson experiments, while sample-movement overhead averaged under 2% of runtime.
- Maintained error below (10^{-2}) on a 10-dimensional PDE and achieved comparable operator-learning accuracy with substantially fewer samples than cited baselines.