|
"Here, we aim to develop surrogate deep learning models to replace finite element (FE) modeling and optimization for characterizing the mechanical response of soft tissues such as brain white matter. The forward problem of predicting the evolution of stress for a given tissue geometry—typically performed using FE modeling—is replaced by a recurrent neural network (RNN) that learns the structure–property relationship between tissue microstructure, material properties and stress-strain behavior. The trained RNN emulates the forward FEM model, significantly accelerating the prediction of the stress-strain response of the microstructure.
By design, these flexible and high-capacity models are able to capture nonlinear and multiscale behavior present in complex brain tissue. However, training ML-based surrogates for scientific applications often faces the challenges of limited, noisy, or biased data from high-fidelity simulations. Here we will further investigate uncertainty quantification methods in both training these models and generating trustworthy predictions."
|
Sign in
to view more information about this project.
