NDSS 2025 – Revisiting Concept Drift In Windows Malware Detection

A session summary highlights a research paper proposing a method to detect and classify drifted Windows malware by learning drift-invariant features in control flow graphs via graph neural networks and adversarial domain adaptation. Compared against active learning retraining and vision-domain adaptation methods, the approach reportedly improves detection on public benchmarks and real-world malware datasets, including multiple drifting malware families, and is presented in the context of the NDSS Symposium 2025.