2021/09/03 by Maarja Bussov, Bussov, Maarja, Joonas Nättilä +1 · 1 citation
Computer Science · Engineering · #Adversarial Robustness in Machine Learning #Anomaly Detection Techniques and Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Physical sciences #Fluid Dynamics and Turbulent Flows #Generative Adversarial Networks and Image Synthesis #High Energy Astrophysical Phenomena (astro-ph.HE) #Machine Learning (cs.LG) #Plasma Physics (physics.plasm-ph)
paper · pdf · doi:10.48550/arxiv.2109.01381
openalex publication_date 2021/09/03 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Computer vision and machine learning tools offer an exciting new way for\nautomatically analyzing and categorizing information from complex computer\nsimulations. Here we design an ensemble machine learning framework that can\nindependently and robustly categorize and dissect simulation data output\ncontents of turbulent flow patterns into distinct structure catalogues. The\nsegmentation is performed using an unsupervised clustering algorithm, which\nsegments physical structures by grouping together similar pixels in simulation\nimages. The accuracy and robustness of the resulting segment region boundaries\nare enhanced by combining information from multiple simultaneously-evaluated\nclustering operations. The stacking of object segmentation evaluations is\nperformed using image mask combination operations. This statistically-combined\nensemble (SCE) of different cluster masks allows us to construct cluster\nreliability metrics for each pixel and for the associated segments without any\nprior user input. By comparing the similarity of different cluster occurrences\nin the ensemble, we can also assess the optimal number of clusters needed to\ndescribe the data. Furthermore, by relying on ensemble-averaged spatial segment\nregion boundaries, the SCE method enables reconstruction of more accurate and\nrobust region of interest (ROI) boundaries for the different image data\nclusters. We apply the SCE algorithm to 2-dimensional simulation data snapshots\nof magnetically-dominated fully-kinetic turbulent plasma flows where accurate\nROI boundaries are needed for geometrical measurements of intermittent flow\nstructures known as current sheets.\n