The integration of deep learning with computational wave imaging (CWI) is rapidly evolving, with a notable shift towards physics-guided neural networks. These networks are being employed to address complex inverse problems in various fields, such as subsurface imaging, flooding area detection, and electromagnetic inversion. A significant development is the discovery of shared hidden properties in the latent spaces of different inverse problems, which suggests a unified approach to solving these problems. This approach leverages the inherent mathematical relationships within the data, enhancing the generalization capabilities of deep learning models. Notably, the use of latent space translations in forward and inverse problems is emerging as a powerful technique, enabling more efficient and accurate solutions. Additionally, the incorporation of physical principles into neural network architectures is leading to more robust and interpretable models, particularly in applications like flooding area detection using SAR imagery. These advancements are paving the way for more sophisticated and versatile computational imaging techniques, capable of handling a broader range of real-world scenarios.

Noteworthy papers include one that demonstrates a hidden property shared by different inverse problems in computational imaging, suggesting a unified approach to these problems. Another paper proposes a physics-guided neural network for flooding area detection, achieving high accuracy and robustness in real-world scenarios.