Data Driven Surrogate

Data-driven surrogate models aim to replace computationally expensive simulations of complex systems with faster, machine-learning-based approximations. Current research emphasizes developing accurate and efficient surrogates using various architectures, including neural networks (e.g., convolutional LSTMs, DeepONets, Graph Neural Networks), Gaussian processes, and polynomial chaos expansions, often incorporating physical constraints or multi-fidelity data for improved performance. These surrogates are proving valuable across diverse fields, accelerating tasks such as weather forecasting, materials science design, and engineering optimization by significantly reducing computational costs and enabling faster design iterations and uncertainty quantification.