A multi-timescale Bayesian optimization method using a neural dynamics model as a Gaussian process prior selected electron cyclotron heating profiles that avoided tearing instabilities in 4 of 8 live DIII-D tokamak shots, versus a 23% historical success rate.
Full shot predictions for the diii-d tokamak via deep recurrent networks
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Multi-Timescale Dynamics Model Bayesian Optimization for Plasma Stabilization in Tokamaks
A multi-timescale Bayesian optimization method using a neural dynamics model as a Gaussian process prior selected electron cyclotron heating profiles that avoided tearing instabilities in 4 of 8 live DIII-D tokamak shots, versus a 23% historical success rate.