Resources for UPT Tutorials

[Project Page] [Paper (arxiv)]]

We recommend to familiarize yourself with the following papers if you haven't already:

• Transformer
• Vision Transformer (ViT)
• Perceiver IO

Tutorials

The following notebooks should provide a tutorial to get familiar with universal physics transformers.

• Preliminaries: provides an introduction into the concepts behind UPT
  • Sparse tensors
  • Positional encoding
  • Architecture overview
• CIFAR10 image classification: start from a basic example (regular grid input, scalar output, easy encoder, simple classification decoder)
• CIFAR10 autoencoder: introduce the perceiver decoder to query at arbitrary positions
• SparseCIFAR10 image classification: introduce handling point clouds via sparse tensors and supernode message passing
• SparseCIFAR10 image autoencoder: combine the handling of input point clouds with decoding arbitrary many positions
• Simple Transient Flow: put everything together to train UPT on a single trajectory of our transient flow simulations