Label Propagation Training Schemes for Physics-Informed Neural Networks and Gaussian Processes
This addresses a specific failure mode in physics-informed machine learning for scientific computing applications, representing an incremental improvement.
The paper tackles the problem of propagating information forward in time in physics-informed machine learning, which is a common failure mode, by proposing semi-supervised training schemes including self-training and co-training for physics-informed neural networks and Gaussian processes, demonstrating their effectiveness through extensive numerical experiments.
This paper proposes a semi-supervised methodology for training physics-informed machine learning methods. This includes self-training of physics-informed neural networks and physics-informed Gaussian processes in isolation, and the integration of the two via co-training. We demonstrate via extensive numerical experiments how these methods can ameliorate the issue of propagating information forward in time, which is a common failure mode of physics-informed machine learning.