PINNfluence: Interpreting PINNs through Influence Functions
Researchers have developed PINNfluence, a new framework designed to interpret the behavior of physics-informed neural networks (PINNs). This method utilizes influence functions to attribute the network's predictions and loss components to specific training data points. Experiments show that PINNfluence can distinguish between well-trained and poorly-trained PINNs, offering a more granular diagnostic tool for understanding and improving their reliability. AI
IMPACT Provides a new method for diagnosing and improving the reliability of physics-informed neural networks used in scientific modeling.