Knowledge Integration into deep learning in dynamical systems: an overview and taxonomy

Despite the sudden rise of AI, it still leaves a question mark to many newcomers on its widespread adoption as it exhibits a lack of robustness and interpretability. For instance, the insufficient amount of training data usually hinders its performance due to the lack of generalization, and the black box nature of deep neural networks does not allow for a precise explanation behind its mechanism preventing a new scientific discovery. Such limitations have led to the development of several branches of deep learning one of which include physics-informed neural networks that will be covered in the rest of this paper. In this overview, we defined the general concept of informed deep learning followed by an extensive literature survey in the field of dynamical systems. We hope to make a contribution to our mechanical engineering community by conveying knowledge and insights on this emerging field of study through this survey paper.

Automated summary

Despite the rapid development of AI, limitations such as lack of robustness and interpretability have hindered its widespread adoption. To overcome these limitations, new branches of deep learning such as physics-informed neural networks have emerged, providing new opportunities for advancement in the field.