Learning Based Surrogate

Learning-based surrogate models aim to replace computationally expensive simulations, such as those solving partial differential equations (PDEs), with faster, data-driven approximations. Current research focuses on improving accuracy and uncertainty quantification in these surrogates, employing diverse architectures like deep neural networks, graph neural networks, and support vector regression, often combined with techniques like proper orthogonal decomposition and active learning to optimize data usage. This accelerates design optimization, enables real-time analysis in applications like digital twins, and facilitates efficient exploration of complex physical systems across various scientific and engineering domains.