An example implementation of a three-dimensional (3D) Vector-Quantized Variational Autoencoder (VQ-VAE) prototype, here used for the compression task of 3D data cubes. This 3D VQ-VAE is an extension of the 2D version developed by airalcorn2. The model comprises of ResNet Encoder and Decoder modules, as well as the Vector Quantization module at the bottleneck. The main motivation for using a VQ-VAE for this compression task is that the vector quantization should produce efficient compressions due to the sparsity in these data. The example 3D data cube used here is a 3D Velocity Distribution Function (VDF) simulated by Vlasiator.
Vlasiator @palmroth2018 is an open-source simulation software used to model the behavior of plasma in the Earth's magnetosphere, a region of space where the solar wind interacts with the Earth’s magnetic field. Vlasiator models collisionless space plasma dynamics by solving the 6-dimensional Vlasov equation, using a hybrid-Vlasov approach. It uses a 3D Cartesian grid in real space, with each cell storing another 3D Cartesian grid in velocity space. The 3D VDF cube we use as an example here is the representation of a single cell in the velocity space. Here's an example of the magnetospheric simulation produced by Vlasiator (credits to Markku Alho and Kostis Papadakis for the following visualization).
