Scale-Split Neural Operator for Memory- and Data-Efficient 3D Turbulence Prediction
My rating
This paper introduces ScaleSplit-NO, a neural surrogate approach for predicting three-dimensional turbulence without relying on full-field numerical simulations at every training resolution. The method exploits the scale structure of turbulence using two neural operators, aiming to reduce memory consumption and the need for expensive full-resolution training data.
Tags: