SIM: Subspace Interaction-based Method for Token-Level Text Anomaly Detection
A hard pseudo-anomaly generation module to construct pseudo-anomalous tokens, simulating the subtle anomalies obscured by semantic smoothing, and a probabilistic boundary loss is developed to standardize anomaly scores into statistical distances, effectively enforcing anomalous instances to deviate significantly from t...
Ke Yan, Yue Tan, Qing-Feng Chen et al.
· 0 citations