Label Smoothed Embedding Hypothesis for Out-of-Distribution Detection
This work provides an incremental improvement in OOD detection for practitioners using classification models, particularly those trained with label smoothing.
This paper proposes an unsupervised method for out-of-distribution (OOD) detection using k-NN density estimation on intermediate activations of a classification model. The method leverages the "Label Smoothed Embedding Hypothesis" and demonstrates improved OOD detection performance both theoretically and empirically when the model is trained with label smoothing, outperforming many existing OOD baselines.
Detecting out-of-distribution (OOD) examples is critical in many applications. We propose an unsupervised method to detect OOD samples using a $k$-NN density estimate with respect to a classification model's intermediate activations on in-distribution samples. We leverage a recent insight about label smoothing, which we call the \emph{Label Smoothed Embedding Hypothesis}, and show that one of the implications is that the $k$-NN density estimator performs better as an OOD detection method both theoretically and empirically when the model is trained with label smoothing. Finally, we show that our proposal outperforms many OOD baselines and also provide new finite-sample high-probability statistical results for $k$-NN density estimation's ability to detect OOD examples.