Classification and dimensionality reduction of international tokamak confinement data on a probabilistic manifold

Pattern recognition for fusion data greatly contributes to a better understanding of the measurements and the physics of fusion plasmas. Through a geometric description of probability it is shown that consideration of the inherent uncertain nature of the data significantly improves the visualization of global confinement data and the identification of confinement regimes. The framework can be extended to the development of scaling laws for ITER.