We are pleased to announce the public release of UNN4AD, the reference implementation of "Can Untrained Neural Networks Detect Anomalies?" (IEEE TII, 2024) by Prof. Seunghyoung Ryu.
UNN4AD scores anomalies by the Mahalanobis distance in the nonlinear random feature space of an untrained neural network, providing a natural performance baseline for tabular anomaly detection. The package exposes a scikit-learn–style API and ships with an ADBench benchmark script for direct comparison.
•
License: PolyForm Noncommercial 1.0.0 (academic & research use)