Neural networks weights related to "Recurrent patterns as a basis for two-dimensional turbulence: predicting statistics from structures"

The dataset contains neural network weights (checkpointed in TensorFlow) for two deep-convolutional autoencoders designed to generate low-dimensional representations of snapshots of vorticity in two-dimensional turbulence. The models have the same architecture apart from the size of the inner-most "embedding" layer. Code to construct the model architecture is also included as a python script.

For details of loss function and training protocol please see associated publication "Recurrent patterns as a basis for two-dimensional turbulence: predicting statistics from structures" (accepted in PNAS, 2024)