Language models improve fusion experiment predictions by focusing on key physics signals
Decomposition-Guided Diffusion Language Models for Inertial Confinement Fusion Prediction
Machine Learning
Summary
Inertial confinement fusion experiments aim to create clean energy but each test costs a lot. Predicting important signals from these experiments is tricky because the key events happen very quickly and past data is limited. The authors made a new method that breaks down the prediction into simpler parts and uses a special type of AI called a language model to improve timing accuracy. Their approach forecasts signals better than previous AI methods on both simulated and real data. This method could help in other science fields where data is scarce and events are rare.
What this means in practice
- •For fusion experiment analysts: Predict neutron rate waveforms from laser pulse and target info more accurately to reduce costly experimental iterations.
- •For industrial process control teams: Improve forecasting of sparse, sensitive signals in manufacturing processes with limited historical data using physics-guided decomposed predictions.
Authors
Xiang Zhang, Varchas Gopalaswamy, Rahman Ejaz, Riccardo Betti, Dongfang Liu
Abstract
Inertial confinement fusion (ICF) is a leading pathway toward clean energy, but each shot at the National Ignition Facility costs on the order of one million dollars, making accurate AI surrogates a high-value target. We study exogenous-driven ICF waveform prediction, where a 512-step neutron-rate diagnostic must be inferred directly from a laser pulse and target design parameters, with no historical response observed. The regime stresses standard time-series predictors with temporal sparsity (picosecond peak in a nanosecond window), input-output scale mismatch (under 300 real shots), and peak sensitivity (picosecond timing). We propose ICF-DLM, to our knowledge the first LM-based ICF predictor, combining (i) a physics-typed decomposition into yield $Y_{DT}$, peak timing $t_{\mathrm{peak}}$, and local waveform $w_{\mathrm{local}}$; (ii) bidirectional denoising that defers commitment to peak location; and (iii) a physics-driven PPO reward re-injecting metric structure across numeric tokens. On ICFBench (50K simulations + 232 experimental shots), ICF-DLM cuts peak-timing error from 11.6 to 9.2 steps over a matched autoregressive LLaMA-3-8B and outperforms classical sequence models and LLM-based time-series predictors. Beyond ICF, the recipe shows potential to address science domains with low data and sparse events.