Computer Science > Machine Learning
[Submitted on 15 Aug 2019 (v1), revised 26 Nov 2020 (this version, v10), latest version 24 May 2021 (v13)]
Title:Entropic Out-of-Distribution Detection
View PDFAbstract:In this paper, we argue that the unsatisfactory out-of-distribution (OOD) detection performance of neural networks is mainly due to the SoftMax loss anisotropy and propensity to produce low entropy probability distributions in disagreement with the principle of maximum entropy. Current out-of-distribution (OOD) detection approaches usually do not directly fix the SoftMax loss drawbacks but rather build techniques to circumvent it. Unfortunately, those methods usually produce undesired side effects (e.g., additional hyperparameters, slower inferences, and collecting extra data). In the opposite direction, we propose replacing SoftMax loss with a novel loss function that does not suffer from the mentioned weaknesses. The proposed IsoMax loss is isotropic (exclusively distance-based) and provides high entropy posterior probability distributions. Replacing the SoftMax loss by IsoMax loss requires no model or training changes. Additionally, the models trained with IsoMax loss produce as fast and energy-efficient inferences as those trained using SoftMax loss. Further, no classification accuracy drop is observed. The proposed method does not rely on outlier/background data, hyperparameter tuning, temperature calibration, feature extraction, metric learning, adversarial training, ensemble procedures, or generative models. Our experiments showed that IsoMax loss works as a seamless SoftMax loss drop-in replacement that significantly improves neural networks OOD detection performance. Therefore, it may be used as a baseline solution to be combined with current or future OOD detection techniques to achieve even higher results.
Submission history
From: David Macêdo [view email][v1] Thu, 15 Aug 2019 14:54:52 UTC (561 KB)
[v2] Sat, 17 Aug 2019 02:22:46 UTC (536 KB)
[v3] Tue, 20 Aug 2019 00:58:03 UTC (573 KB)
[v4] Wed, 5 Feb 2020 01:18:56 UTC (279 KB)
[v5] Tue, 18 Feb 2020 23:06:49 UTC (232 KB)
[v6] Sun, 7 Jun 2020 06:54:51 UTC (649 KB)
[v7] Thu, 11 Jun 2020 05:59:13 UTC (647 KB)
[v8] Thu, 23 Jul 2020 17:03:57 UTC (1,118 KB)
[v9] Mon, 3 Aug 2020 06:37:15 UTC (261 KB)
[v10] Thu, 26 Nov 2020 22:47:49 UTC (3,836 KB)
[v11] Mon, 30 Nov 2020 01:55:08 UTC (3,836 KB)
[v12] Sun, 11 Apr 2021 23:32:49 UTC (4,067 KB)
[v13] Mon, 24 May 2021 23:15:23 UTC (963 KB)
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
IArxiv Recommender
(What is IArxiv?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.