Home

Awesome

Self-Supervised Learning for Graph Dataset Condensation

This is the Pytorch implementation of KDD'24 work: "Self-Supervised Learning for Graph Dataset Condensation".

Requirements

Instructions

SGDC

We provide the shell scripts in "command.sh" to run the code on NCI1 and ogbg-molhiv dataset.

Kindly note, the file "pretrain" includes the target embeddings, which are generated by GraphCL. So you can generate any target embeddings for your own datasets.

SGDC-Att

The code of attention-based neural network kernel is in "antk.py", you can change the graph kernel in model training (line 210 in main.py).

SGDC-CL

We also implement the code for initial design, SGDC-CL, in "UnspervisedGM/" folder.