Physic Informed Gaussian Process

Physics-informed Gaussian processes (PIGPs) combine the probabilistic framework of Gaussian processes with the constraints imposed by known physical laws, primarily expressed as partial differential equations (PDEs). Research focuses on developing efficient algorithms, such as those employing Markov Chain Monte Carlo methods or variational autoencoders, to infer model parameters and predict system behavior from limited data, often incorporating active learning strategies for optimal data acquisition. This approach enhances the accuracy and reliability of predictions in various engineering applications, including structural health monitoring and topology optimization, by leveraging both data and physical understanding to reduce uncertainty and improve model interpretability. The resulting models offer a powerful tool for uncertainty quantification and improved decision-making in complex systems.