A Variational Dirichlet Framework for Out-of-Distribution Detection

Sep 27, 2018 ICLR 2019 Conference Blind Submission readers: everyone Show Bibtex
  • Abstract: With the recently rapid development in deep learning, deep neural networks have been widely adopted in many real-life applications. However, deep neural networks are also known to have very little control over its uncertainty for test examples, which potentially causes very harmful and annoying consequences in practical scenarios. In this paper, we are particularly interested in designing a higher-order uncertainty metric for deep neural networks and investigate its performance on the out-of-distribution detection task proposed by~\cite{hendrycks2016baseline}. Our method first assumes there exists a underlying higher-order distribution $\mathcal{P}(z)$, which generated label-wise distribution $\mathcal{P}(y)$ over classes on the K-dimension simplex, and then approximate such higher-order distribution via parameterized posterior function $p_{\theta}(z|x)$ under variational inference framework, finally we use the entropy of learned posterior distribution $p_{\theta}(z|x)$ as uncertainty measure to detect out-of-distribution examples. However, we identify the overwhelming over-concentration issue in such a framework, which greatly hinders the detection performance. Therefore, we further design a log-smoothing function to alleviate such issue to greatly increase the robustness of the proposed entropy-based uncertainty measure. Through comprehensive experiments on various datasets and architectures, our proposed variational Dirichlet framework with entropy-based uncertainty measure is consistently observed to yield significant improvements over many baseline systems.
  • Keywords: out-of-distribution detection, variational inference, Dirichlet distribution, deep learning, uncertainty measure
  • TL;DR: A new framework based variational inference for out-of-distribution detection
0 Replies

Loading