Appendix
–Neural Information Processing Systems
In the finetuning process, we consider 8 benchmark binary classification datasets from MoleculeNet. The detailed statistics of datasets are listed in Table 1. Here, we show the pseudo code of the multi-level self-supervised pre-training (Algorithm 1). The output is the model parameters of pre-trained GNNs. In each training iteration, we update the parameters of all the pretext models.
Neural Information Processing Systems
Aug-15-2025, 15:14:29 GMT