Skip to main content

Towards Expressive Graph Neural Networks Beyond Message-Passing

In this project, we develop expressive neural network architectures for learning graphs.

Expressive Pure Graph Transformers without Message-Passing​

Graph Inductive Biases in Transformers without Message Passing (ICML 2023)

SOTA in 2023! Still Competitive Now!

PWC PWC PWC PWC PWC PWC PWC

TL;DR​

  • Propose a novel graph positional encoding scheme: Relative Random Walk Probabilities (RRWP)
  • Propose graph transformers architecture without MPNNs. (Pure Transformer! Not just a painkiller for MPNNs!)
  • Beat all previous graph transformers with/without MPNNs.

drawing

drawing

Graph Convolution on Peusdo-coordinates​

CKGConv: General Graph Convolution with Continuous Kernels (ICML 2024)

Team​