Brief

Researchers have developed two inverse frameworks, a differentiable physics approach and physics-informed neural networks (PINNs), to reconstruct wall shear stress (WSS) from passive scalar observations. The study evaluated these methods on canonical and patient-specific cardiovascular flow problems. The differentiable physics framework demonstrated superior accuracy and robustness across various measurement scenarios compared to PINNs, particularly when near-wall data was limited. AI

Researchers have introduced DIfferentiable Autoencoding Neural Operator (DIANO), a novel framework designed to create interpretable and computationally efficient latent spaces for scientific machine learning. DIANO utilizes neural operators for both dimensionality reduction and reconstruction, enabling the enforcement of physical laws directly within the latent space. This approach has demonstrated accurate reconstruction of complex spatiotemporal data across various benchmark problems, including fluid dynamics simulations, at a reduced computational cost. AI