Reproduce “Memorization in Graph Neural Networks” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Adarsh Jamadandi, Jing Xu, Adam Dziedzic, Franziska Boenisch

NCMemo measures how seeing a node’s label during training changes a graph neural network’s prediction for that node. It compares matched models trained with different candidate nodes and counts scores above 0.5 as memorization. In synthetic syn-cora, a three-layer GCN’s memorization rate falls from 83.63% at homophily 0.0 to 0% at 1.0; rewiring and privacy results remain synthetic.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.