SERGCL (Stochastic Experience-Replay for Graph Continual Learning)

This is the code for the paper Stochastic Experience-Replay for Graph Continual Learning.

The following figure compares the typical sampling-based replay for GCL, condensation-based replay for GCL, and stochastic condensation-based replay (proposed method) for GCL.

The following figure presents an illustration of SERGCL.

Main experiment results.

Experiment environment

Our experiments are run on the enviroment based on Python 3.11.5 the rest of the packages used to conduct the experiments can be found in requirements.yml

Usage

To reproduce the results of Table II (classIL setting) and Table III (taskIL setting), please execute run.sh.

Credit

This repository was developed based on the following two repositories: CaT-CGL and CGLB.