Covariance Corrected Mahalanobis Distance for OOD Intent Detection
September 2, 2026
A new covariance corrected Mahalanobis distance method improves out-of-domain (OOD) intent detection for conversational agents. This approach specifically addresses performance failures observed in previous Mahalanobis-based methods when applied to few-shot settings.
HOW THIS AFFECTS YOU
●
builderYou can improve the robustness of chatbots when encountering unfamiliar user intents.
●
researcherThis provides a mathematical correction to improve few-shot OOD classification performance.