Plasma Kink Classification Using Deep Learning

Contacts: mteng-levy@lanl.gov, bwolfe@lanl.gov

Many types of plasma instabilities are observed in laboratory plasma experiments. Even though the fundamental mechanisms are known, many phenomena and features of plasma instabilities are too complex to be fully describable by theory or even simulations. For example, in plasma jets, a rich variety of plasma kinks can arise, which differ in kink amplitudes, radial acceleration, temporal evolution, transition to Rayleigh-Taylor instability, and break-away of the plasma jet. Classification of plasma kinks using machine learning offers a new approach to study and understand the complexity of plasma kinks. Furthermore, accumulative plasma movies from Caltech offer a sufficiently large amount of data for this work [1]. We adopt deep neutral network classification methods such as alexnet, Resnet18 for the plasma kink image classification workflow. In biological terms, our longer term goal of classifying plasma kinks is to connect phenotypical features from the images to genotypical or physical interpretation of the observations.

 

[1] You, S., Yun, G. S., & Bellan, P. M. 2005, PhRvL, 95, 04500.