Mdlvzip

: Users can select different bottleneck sizes to balance how much space they save versus how accurate the reconstruction is.

: You can share compact "latent representations" instead of entire raw trajectories, making collaboration easier. Mdlvzip

: The encoder takes the complex atomic positions and maps them into a tiny "bottleneck" size. : Users can select different bottleneck sizes to

: It doesn't need complex physics rules during the compression phase, making it faster and more flexible. Mdlvzip

The core of MDZip is a neural network called a .

: It shrinks data by more than 95% compared to raw formats.