Explainable Meta Bayesian Optimization with Human Feedback for Fusion Energy

Fusion energy offers the potential for unlimited clean energy, crucial for a sustainable future. Inertial Confinement Fusion (ICF) experiments are costly and limited in frequency, necessitating sample-efficient optimization techniques. Traditional Bayesian Optimization (BO) methods face limitations in this context. This paper introduces Meta Bayesian Optimization with Human Feedback (MBO-HF), which integrates Meta-Learning and expert preferences to enhance BO. MBO-HF employs Transformer Neural Processes (TNPs) to create a meta-learned surrogate model and a human-informed acquisition function to suggest and explain candidate experiments. MBO-HF outperforms current methods in optimizing the energy yield of the ICF and is effective in benchmarking tasks.