Predicting Areal Density in ICF Experiments Using an Uncertainty-Aware Neural Network

The areal density is a key performance metric in achieving ignition and high gains in ICF. Accurately predicting areal density is therefore critical for optimizing these experiments. This study develops a machine learning model designed to accurately predict the areal density in experiments using data from cryogenic direct-drive experiments on OMEGA. The input parameters for the model include variables related to the target design and laser pulse conditions and use the radiation-hydrodynamic code LILAC to provide physically consistent embeddings. Uncertainty quantification is also incorporated, adding confidence to predictions and supporting informed experimental design and analysis. While current models provide reasonable estimates and are commonly used at the LLE, exploring machine learning offers the potential for improved predictive performance and uncertainty estimation. This material is based upon work supported by the Department of Energy [National Nuclear Security Administration] University of Rochester “National Inertial Confinement Fusion Program” under Award Number(s) DE-NA0004144.