Cross-Machine Transfer Learning for a Compact Spherical-Tokamak Breeder: A Disruption Model Pre-Trained on TCV, DIII-D, and NSTX with Honest Transfer Uncertainty
Abstract
A first-of-a-kind fusion breeder has no operating record of its own, so its disruption-prediction model must be born from other machines' data — and it must know how far it is being asked to reach. We formalise cross-machine transfer learning for the Hyperion compact spherical-tokamak breeder (frozen design point I_p=9.66, Q=3.076, P_ fus=85.04, negative triangularity δ=-0.30, B_0=8, elongation κ=2.0, major radius =1.20, aspect ratio A=2.5, centrepost peak field 16.84; configuration 22021). A disruption model is pre-trained on the public-tokamak corpus — TCV, DIII-D and NSTX — and transferred to the frozen breeder configuration through a governed bridge. We cast the problem formally: disruption warning as a discrete-time hazard, transfer as domain adaptation under covariate and prior shift, and the reachable target risk as the Ben-David bound _T(h)_S(h)+12(,)+. Feature alignment (maximum-mean-discrepancy penalty, second-order CORAL and importance weighting) shrinks the divergence term; split-conformal prediction under covariate shift makes the model's uncertainty honest with a distribution-free coverage guarantee; and a machine-readable extrapolation ledger (register anchor KX-L1-A7, a resolved nine-device post-diction spanning the breeder and burner) supplies a support functional that caps model authority where cross-machine support is thin. On a deterministic, reproducible synthetic transfer study built to the frozen ledger, pre-training plus fine-tuning lifts held-out…
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.