FusionMMT: A Unified Multimodal and Multitask Learning Framework for Nuclear Fusion
FusionMMT is presented, the first unified multimodal multitask framework for intelligent tokamak plasma diagnostics that combines task-adaptive multimodal fusion with progressive multitask optimization to learn shared and task-specific representations while mitigating cross-task conflicts and optimization imbalance.
Qiang Chen, Xiao Wang, Qing-Quan Yang et al.
· 0 citations