Moving beyond testing on in-distribution data works on Out-of-Distribution (OOD) detection have recently increased in popularity. A recent attempt to categorize OOD data introduces the concept of near and far OOD detection. Specifically, prior works define characteristics of OOD data in terms of detection difficulty. We propose to characterize the spectrum of OOD data using two types of distribution shifts: covariate shift and concept shift, where covariate shift corresponds to change in style, e.g., noise, and concept shift indicates a change in semantics. This characterization reveals that sensitivity to each type of shift is important to the detection and confidence calibration of OOD data. Consequently, we investigate score functions that capture sensitivity to each type of dataset shift and methods that improve them. To this end, we theoretically derive two score functions for OOD detection, the covariate shift score and concept shift score, based on the decomposition of KL-divergence for both scores, and propose a geometrically-inspired method (Geometric ODIN) to improve OOD detection under both shifts with only in-distribution data. Additionally, the proposed method naturally leads to an expressive post-hoc calibration function which yields state-of-the-art calibration performance on both in-distribution and out-of-distribution data. We are the first to propose a method that works well across both OOD detection and calibration and under different types of shifts. View project page at https://sites.google.com/view/geometric-decomposition.
翻译:最近试图将OOOD数据分类,引入了近距离和远距离OOOD检测的概念。具体地说,先前的工作界定OOOD数据在检测难度方面的特性。我们提议使用两种分布变化类型来描述OOD数据的频谱:共变转移和概念转变,即共变转移与风格变化相对应,例如噪音和概念转变表明语义变化。这种定性表明,对每类转移的敏感性对于OOOD数据的检测和信心校准十分重要。因此,我们调查的分数功能能够捕捉到对每类数据集变化的敏感性以及改进数据的方法。为此,我们从理论上为OOOD的检测、共变换变换分和概念转变评分得出两个评分功能,根据KL-divorgence对分的变换位,提出一种对地校正判法方法(ODING)来改进OOD的检测工作,但只是在分配期间进行这种变换和变校准后,因此,拟议的方法自然得出了O-ral-ral-ral-ral-deal-deal-deal-deal-laction laction laction-laction-laction-laction-laction-laction-lade-lade-lade-de laction-labal laction-laction-laction-labal laction-de-de-de-de-laut laction-de-de-de-de-de-de-de-de-de-de-de-la-laction-laction-laction-laction-laction-laction-laction-laction-laction-de-de-de-laction-laction-laction-laction-de-de-de-de-de-de-laction-de-de-de-de-de-de-de-de-de-laction-de-laction-laction-laction-laction-laction-laction-de-de-de-de-lad-laction-laction-de-de-de-de-de-laction-de-laction-laction-de-laction-lad-la-la-la-la-la-la-la-la-la-