Rethinking the impact of noisy labels in graph classification: A utility and privacy perspective

This is the code implementation of the method from the paper "Rethinking the impact of noisy labels in graph classification: A utility and privacy perspective." arXiv:2406.07314

Requirements

Hardware environment: NVIDIA Tesla V100-SXM2-32GB chip with 32GB memory.
Software environment: python 3.8, CUDA 11.3, Ubuntu 18.04.6 and PyTorch 1.12.1

Training

Here, we take NCI1 as an example:

python RgLc.py --dataset NCI1 --fold_idx 0 --ptb 0.3