Optimizing Three Critical Factors for Practical and Effective OOD Detection Fine-Tuning
The episode discusses a paper by researchers from Seoul National University and RideFlux regarding Out-of-Distribution (OOD) detection. The hosts explore how to resolve the trade-off between classification accuracy and OOD detection using three specific methods: Self-Knowledge Distillation, semi-hard outlier sampling, and Outlier-aware Supervised Contrastive Learning to improve model reliability.