Developing deep learning algorithms for inferring upstream separatrix density at JET

Predictive and real-time inference capability for the upstream separatrix electron density, $n_\text{e, sep}$, is essential design control of core-edge integrated plasma scenarios. In this study, both supervised semi-supervised machine learning algorithms are explored to establish direct mapping as well indirect compressed representation pedestal profiles predictions $n_{\text{e, sep}}$. Based on EUROfusion database JET, a tabular dataset was created, consisting parameters, fraction ELM cycle, high resolution Thomson scattering density temperature, sep}}$ 608 JET shots. Using dataset, approach provides parameters percentage Through learning, experimental temperature established. By conditioning with probabilistic generative predictive model For prediction, can be used conditional distribution profiles, decoder that trained part algorithm decode back full profiles. Although, in work, proof-of-principle predicting inferring given, such also many other applications profile predicted. An implementation work found at https://github.com/fusionby2030/moxie.