RankOOD -- Class Ranking-based Out-of-Distribution Detection

TL;DR

RankOOD introduces a novel rank-based out-of-distribution detection method using Placket-Luce loss, achieving state-of-the-art results on TinyImageNet by leveraging class ranking patterns.

Contribution

It formalizes class ranking patterns with Placket-Luce loss for improved OOD detection, a novel application in this context.

Findings

Achieves state-of-the-art performance on TinyImageNet OOD detection benchmark.

Reduces FPR95 by 4.3% compared to previous methods.

Utilizes class ranking patterns to distinguish in-distribution from OOD samples.

Abstract

We propose RankOOD, a rank-based Out-of-Distribution (OOD) detection approach based on training a model with the Placket-Luce loss, which is now extensively used for preference alignment tasks in foundational models. Our approach is based on the insight that with a deep learning model trained using the Cross Entropy Loss, in-distribution (ID) class prediction induces a ranking pattern for each ID class prediction. The RankOOD framework formalizes the insight by first extracting a rank list for each class using an initial classifier and then uses another round of training with the Plackett-Luce loss, where the class rank, a fixed permutation for each class, is the predicted variable. An OOD example may get assigned with high probability to an ID example, but the probability of it respecting the ranking classification is likely to be small. RankOOD, achieves SOTA performance on the near-ODD TinyImageNet evaluation benchmark, reducing FPR95 by 4.3%.