Stabilizing PINNs: A regularization scheme for PINN training to avoid unstable fixed points of dynamical systems
This addresses a specific training instability issue for PINNs in solving dynamical systems, which is an incremental but practical improvement for researchers and practitioners in scientific machine learning.
The paper tackles the problem of physics-informed neural networks (PINNs) getting stuck at unstable fixed points when solving initial value problems, which leads to physically incorrect solutions. The authors propose a regularization scheme that penalizes unstable fixed points, and experimental results on four dynamical systems show it substantially improves training success rates.
It was recently shown that the loss function used for training physics-informed neural networks (PINNs) exhibits local minima at solutions corresponding to fixed points of dynamical systems. In the forward setting, where the PINN is trained to solve initial value problems, these local minima can interfere with training and potentially leading to physically incorrect solutions. Building on stability theory, this paper proposes a regularization scheme that penalizes solutions corresponding to unstable fixed points. Experimental results on four dynamical systems, including the Lotka-Volterra model and the van der Pol oscillator, show that our scheme helps avoiding physically incorrect solutions and substantially improves the training success rate of PINNs.