We propose black-box tests to detect harmful memorization in foundation models trained on structured EHR data. Validated on a public model, our toolkit supports privacy audits by distinguishing generalization from privacy-compromising memorization.

This paper proposes a scalable local search method for detecting balanced polarized communities in signed networks, including neutral nodes, using a new objective that avoids size imbalance and achieves superior quality and efficiency.

We present HollowFlow, a flow-based generative model based on a GNN that makes sample likelihood computations drastically more efficient.