Gradient-based optimization of nuclear criticality experiments using neural surrogate eigenvalue sensitivities
The work uses a physics-informed deep neural network to predict how changing an experiment's geometry affects its neutronic similarity ($c_k$) to a target reactor design. By using this differentiable model, the researchers optimized new experiment designs using gradient-based methods to find geometries that maximize $c_k$, successfully generating configurations with high similarity scores for a transportation cask.