The field of physics-informed neural networks (PINNs) is rapidly advancing, with a focus on improving the accuracy and efficiency of solving partial differential equations (PDEs). Recent developments have highlighted the importance of incorporating physical constraints and geometry awareness into neural network architectures. This has led to the development of novel methods, such as the use of liquid residual blocks and measurement-aware cross attention mechanisms, which have shown significant improvements in predictive accuracy and robustness. Notably, the introduction of physics-informed loss functions and geometry-aware spatio-spectral graph neural operators has enabled the solution of complex PDEs with high accuracy and efficiency. Furthermore, the application of PINNs to real-world problems, such as bearing fault classification and satellite attitude dynamics, has demonstrated their potential for practical impact.
Noteworthy papers include: Fast, Convex and Conditioned Network for Multi-Fidelity Vectors and Stiff Univariate Differential Equations, which introduces a simple yet effective activation filtering step to increase matrix rank and expressivity while preserving convexity. Diffeomorphic Neural Operator Learning, which presents an operator learning approach for a class of evolution operators using a composition of a learned lift into the space of diffeomorphisms of the domain and the group action on the field space.