Turbulence is an extremely difficult problem to understand and model. An increasingly popular approach is to use Machine Learning (ML) methods to try to develop models of turbulence that are trained using data. We have done some initial work in this area to develop a model that can perform large compression on turbulent flow data while minimizing the information lost. Such compression models are typically needed for use in conjunction with sequence prediction models when developing ML models capable of predicting the time evolution of a dynamical system. Compression models are also needed to effectively reduce the size of simulation restart files that are needed when performing large-scale simulations of turbulent flows. A representative publication on this work is:
M. Momenifar, E. Diao, V. Tarokh, A.D. Bragg, Dimension Reduced Turbulent Flow Data From Deep Vector Quantizers, Journal of Turbulence, 2021, under review (https://arxiv.org/abs/2103.01074).