Learning Confidence for Out-of-Distribution Detection in Neural Networks
By Terrance Devries, Graham W. Taylor. 2018
664 citations recorded by the source corpus.
Modern neural networks are very powerful predictive models, but they are often incapable of recognizing when their predictions may be wrong. Closely related to this is the task of out-of-distribution detection, where a network must determine whether or not an input is outside of the set on which it is expected to safely perform. To jointly address these issues, we propose a method of learning confidence estimates for neural networks that is simple to implement and produces intuitively interpretable outputs. We demonstrate that on the task of out-of-distribution detection, our technique surpasses recently proposed techniques which construct confidence based on the network's output distribution, without requiring any additional labels or access to out-of-distribution examples. Additionally, we address the problem of calibrating out-of-distribution detectors, where we demonstrate that misclassified in-distribution examples can be used as a proxy for out-of-distribution examples.
Subfields: Artificial Intelligence, Safety, Robustness & Interpretability, Anomaly & Outlier Detection
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
- Mohit Bansal
- Sergey Levine
- Yoshua Bengio
- Percy Liang
- Jianfeng Gao
- Quoc V. Le
- Caiming Xiong
- Richard Socher
In Safety, Robustness & Interpretability
In Anomaly & Outlier Detection
- Yixuan Li
- Yiyou Sun
- Dan Hendrycks
- Mung Chiang
- José Miguel Hernández-Lobato
- Prateek Mittal
- Dawn Song
- Julie Arnold Shah
Citations in the corpus
- Out-of-Domain Detection for Natural Language Understanding in Dialog Systems (2019, neutral): “, 2017), applying Bayesian models (Malinin and Gales, 2018), or even explicitly learning a detection score (DeVries and Taylor, 2018).”
- Deep Anomaly Detection with Outlier Exposure (2018, methodology): “A recently proposed OOD detection technique (DeVries & Taylor, 2018) involves appending an OOD scoring branch b : X → [0, 1] onto a deep network.”
- Large-Scale Long-Tailed Recognition in an Open World (2019, background): “Open-set recognition [44, 2], or out-of-distribution detection [9, 26], aims to re-calibrate the sample confidence in the presence of open classes.”
- Practical Deep Learning with Bayesian Principles (2019, methodology): “We use experimental protocol from the literature [16, 31, 8, 32] to compare VOGN, Adam and MC-dropout on CIFAR-10.”
- Energy-based Out-of-distribution Detection (2020, background): “DeVries and Taylor [6] propose to learn the confidence score by attaching an auxiliary branch to a pre-trained classifier and deriving an OOD score.”
- Scaling Out-of-Distribution Detection for Real-World Settings (2022, methodology): “ODIN uses temperature scaling and adds an epsilon perturbation to the input in order to separate the softmax posteriors between in- and out-of-distribution images; we set these hyperparameters following (DeVries & Taylor, 2018).”
- ProtoAttend: Attention-Based Prototypical Learning (2019, background): “In (DeVries and Taylor, 2018), direct inference of confidence output is considered with a modified loss.”
- Unsupervised Class-Incremental Learning Through Confusion (2021, background): “DeVries & Taylor (2018) incorporates a confidence branch to obtain out-of-distribution estimations.”
- Addressing Failure Prediction by Learning Model Confidence (2019, background): “[6] share with us the same purpose of learning confidence in neural networks.”
- Deterministic Neural Networks with Appropriate Inductive Biases Capture Epistemic and Aleatoric Uncertainty (2021, methodology): “Popular approaches to tackle this problem include pre-processing of inputs and post-hoc calibration methods [Liang et al., 2018, Guo et al., 2017], alternative objective functions [Lee et al., 2018a, DeVries and Taylor, 2018], and exposure to outliers [Hendrycks et al., 2018].”
- Contrastive Training for Improved Out-of-Distribution Detection (2020, methodology): “MSP scoring was improved by using strategies like confidence loss [17], auxiliary objectives [6, 13, 23], margin loss [34] and outlier exposure [12].”
- OpenOOD: Benchmarking Generalized Out-of-Distribution Detection (2022, methodology): “methods take the major part of OOD detectors, which gets confidence directly from the classifier with some design that is beneficial to OOD detection [9, 10].”
- Evaluating Merging Strategies for Sampling-based Uncertainty Techniques in Object Detection (2018, methodology): “2) Uncertainty Error (UE): also called detection error, has previously been used in literature to evaluate uncertainty effectiveness for classification tasks [9], [30].”