Recent work has shown that a much simpler model, simple graph convolution (SGC) (Wu et al., 2019),iscompetitivewithGCNs incommon graph machine learning benchmarks.

However, the cross-client edges naturally exist among clients.

Pseudo labeling (PL) is a wide-applied strategy to enlarge the labeled dataset by self-annotating the potential samples during the training process.

Our study aims to promote a higher standard of empirical rigor in the field of graph machine learning, encouraging more accurate comparisons and evaluations of model capabilities.

In Table 1, we report the cross-task testbed results in two transfer learning settings, i.e., (a) node classification to link prediction (Table 1a) and (b) link prediction to node classification (Table 1b).

These limitations hinder GNNs' ability to capture long-range dependencies in graphs, ultimately