Leveraging Latent Representations for Predictive Physics-Based Modeling

Explore cutting-edge approaches for leveraging latent representations in predictive physics-based modeling through this seminar presented by Katiana Kontolati from Bayer. Delve into innovative techniques that address the challenges of uncertainty quantification (UQ) in complex partial differential equations (PDEs). Learn about a manifold-based approach for probabilistic parameterization of nonlinear PDEs using atomistic simulation data, applied to modeling plastic deformation in bulk metallic glass systems. Discover the Latent Deep Operator Network (L-DeepONet) for training neural operators on latent spaces, enhancing predictive accuracy for time-dependent PDEs. Examine a transfer learning framework based on Hilbert space embeddings of conditional distributions for PDE regression using DeepONet, enabling generalizability and reducing computational resource requirements. Gain insights into the application of these methods in complex physics and engineering problems, presented by an expert in physics-informed machine learning and high-dimensional surrogate modeling.