Fewer Shots, Better Implosions: Sample-Efficient Optimization for Inertial Confinement Fusion

The global demand for clean energy has brought Inertial Confinement Fusion (ICF) to the forefront of sustainable power research. Due to the high cost and limited availability of ICF experiments, optimization methods must achieve exceptional sample efficiency. Bayesian Optimization (BO) is a standard tool for expensive black-box function optimization, yet it suffers from a “cold-start” problem, neglecting prior knowledge from simulations and past experiments. We propose a Meta-Bayesian Optimization (Meta-BO) approach that integrates prior tasks into the BO loop. Our method introduces boundary-box constraints, dual acquisition strategies, and interpretability candidates. Our method (MBO) achieves substantial performance gains over existing BO approaches in ICF energy-yield optimization, demonstrating sample efficiency and accelerating the path to a sustainable fusion power source.