ReAct: Out-of-distribution Detection With Rectified Activations

By Yiyou Sun, Chuan Guo, Yixuan Li. Neural Information Processing Systems, 2021

694 citations recorded by the source corpus.

Out-of-distribution (OOD) detection has received much attention lately due to its practical importance in enhancing the safe deployment of neural networks. One of the primary challenges is that models often produce highly confident predictions on OOD data, which undermines the driving principle in OOD detection that the model should only be confident about in-distribution samples. In this work, we propose ReAct--a simple and effective technique for reducing model overconfidence on OOD data. Our method is motivated by novel analysis on internal activations of neural networks, which displays highly distinctive signature patterns for OOD distributions. Our method can generalize effectively to different network architectures and different OOD detection scores. We empirically demonstrate that ReAct achieves competitive detection performance on a comprehensive suite of benchmark datasets, and give theoretical explication for our method's efficacy. On the ImageNet benchmark, ReAct reduces the false positive rate (FPR95) by 25.05% compared to the previous best method.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/7f9760a76e9cf424da0b72d42f75594cefc4a329

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Generalized Out-of-Distribution Detection: A Survey (2021, positive): “Therefore, ReAct (Sun et al., 2021b) proposes truncating the high activations, which establishes strong post-hoc detection performance and further boosts the performance of existing scoring functions.”
Is Out-of-Distribution Detection Learnable? (2022, methodology): “[17] Yiyou Sun, Chuan Guo, and Yixuan Li.”
ViM: Out-Of-Distribution with Virtual-logit Matching (2022, methodology): “For ReAct, we use the Energy+ReAct setting with rectification percentile p = 99.”
INSIDE: LLMs' Internal States Retain the Power of Hallucination Detection (2024, methodology): “Inspired by prior works that rectify internal activations to reduce overconfident prediction for Out-of-Distribution (OOD) detection (Sun et al., 2021; Djurisic et al., 2022; Chen et al., 2024), we introduce a test time feature clipping ( FC ) method to prevent LLMs generate overconfident…”
Mitigating Neural Network Overconfidence with Logit Normalization (2022, background): “Indeed, many prior works turned to deﬁne alternative OOD scoring functions (Liang et al., 2018; Lee et al., 2018; Liu et al., 2020; Sastry & Oore, 2020; Sun et al., 2021; Huang et al., 2021; Sun et al., 2022).”
CMG: A Class-Mixed Generation Approach to Out-of-Distribution Detection (2022, methodology): “(10) React [52]: This method exploits the internal activations of neural networks to find distinctive signature patterns for OOD distributions.”
OpenOOD: Benchmarking Generalized Out-of-Distribution Detection (2022, background): “Later works explore other simple and more efﬁcient indicators to distinguish ID and OOD, such as ODIN [13] that uses temperature scaling with gradient-based input perturbations, MDS [18] that measures minimum mahalanobis distance from class centroids, EBO [14] that uses energy-based functions, GRAM ”
Out-of-distribution Detection with Deep Nearest Neighbors (2022, methodology): “Our comparison covers an extensive collection of outlier detection methods in literature including: IForest (Liu et al., 2008), OCSVM (Sch ¨ olkopf et al., 2001), LODA (Pevn ` y, 2016), PCA (Shyu et al., 2003), and LOF (Breunig et al., 2000).”