Taming False Positives in Out-of-Distribution Detection with Human Feedback

arXiv:2404.16954v1 Announce Type: new
Abstract: Robustness to out-of-distribution (OOD) samples is crucial for safely deploying machine learning models in the open world. Recent works have focused on designing scoring functions to quantify OOD uncertainty. Setting appropriate thresholds for these scoring functions for OOD detection is challenging as OOD samples are often unavailable up front. Typically, thresholds are set to achieve a desired true positive rate (TPR), e.g., $95\%$ TPR. However, this can lead to very high false positive rates (FPR), …

abstract arxiv cs.ai cs.lg designing detection distribution false false positives feedback functions human human feedback machine machine learning machine learning models robustness samples scoring stat.ml type uncertainty world