优化三个关键因素以实现实用且有效的OOD检测微调
Optimizing Three Critical Factors for Practical and Effective OOD Detection Fine-Tuning
AI总结：
本文通过优化模型提醒、数据采样和表示学习三个关键因素，提出SKD、SOS和OSCL方法，在提升OOD检测性能的同时保持分类准确率，尤其在长尾场景中表现优异。
摘要：
在分布外（OOD）检测中，使用辅助离群数据进行微调通常能提升检测性能，但代价是分类准确率下降。这种权衡源于微调过程中原始分布内（ID）分布的丢失。为了建立更实用且有效的范式，我们优化了三个关键因素：模型提醒、数据采样和表示学习。我们提出：（1）自知识蒸馏（SKD）以缓解准确率下降；（2）半硬离群采样（SOS）以用最少数据提高检测效率；（3）离群感知的监督对比学习（OSCL）以促进ID与OOD的可分离性。优化这些因素产生累积增益，同时提升OOD检测性能和分类准确率。我们的框架在多种基准上优于现有方法，尤其在长尾场景中表现突出，为现实世界的OOD检测提供了稳健的基线。
英文摘要：
In out-of-distribution (OOD) detection, fine-tuning with auxiliary outlier data often improves detection performance at the cost of classification accuracy. This trade-off stems from the loss of the original in-distribution (ID) distribution during fine-tuning. To establish a more practical and effective paradigm, we optimize three critical factors: model reminder, data sampling, and representation learning. We propose: (1) Self-Knowledge Distillation (SKD) to mitigate accuracy reduction; (2) Semi-hard Outlier Sampling (SOS) to improve detection efficiency with minimal data; and (3) Outlier-aware Supervised Contrastive Learning (OSCL) to promote ID-OOD separability. Optimizing these factors produces cumulative gains, boosting both OOD detection performance and classification accuracy. Our framework outperforms existing methods across diverse benchmarks, particularly in long-tailed scenarios, providing a robust baseline for real-world OOD detection.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
