RPM-Net learns reciprocal points for each known attack class plus adversarial constraints to detect unknown threats, with RPM-Net++ adding Fisher regularization, and reports better F1, AUROC, and AUPR-OUT than prior methods.
A baseline for detecting misclassified and out-of-distribution examples in neural networks,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
RPM-Net Reciprocal Point MLP Network for Unknown Network Security Threat Detection
RPM-Net learns reciprocal points for each known attack class plus adversarial constraints to detect unknown threats, with RPM-Net++ adding Fisher regularization, and reports better F1, AUROC, and AUPR-OUT than prior methods.