{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,7]],"date-time":"2026-06-07T07:30:22Z","timestamp":1780817422544,"version":"3.54.1"},"reference-count":27,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2024,11,20]],"date-time":"2024-11-20T00:00:00Z","timestamp":1732060800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Big Data"],"abstract":"<jats:sec><jats:title>Introduction<\/jats:title><jats:p>Multi-layer aggregation is key to the success of out-of-distribution (OOD) detection in deep neural networks. Moreover, in real-time systems, the efficiency of OOD detection is equally important as its effectiveness.<\/jats:p><\/jats:sec><jats:sec><jats:title>Methods<\/jats:title><jats:p>We propose a novel early stopping OOD detection framework for deep neural networks. By attaching multiple OOD detectors to the intermediate layers, this framework can detect OODs early to save computational cost. Additionally, through a layer-adaptive scoring function, it can adaptively select the optimal layer for each OOD based on its complexity, thereby improving OOD detection accuracy.<\/jats:p><\/jats:sec><jats:sec><jats:title>Results<\/jats:title><jats:p>Extensive experiments demonstrate that our proposed framework is robust against OODs of varying complexity. Adopting the early stopping strategy can increase OOD detection efficiency by up to 99.1% while maintaining superior accuracy.<\/jats:p><\/jats:sec><jats:sec><jats:title>Discussion<\/jats:title><jats:p>OODs of varying complexity are better detected at different layers. Leveraging the intrinsic characteristics of inputs encoded in the intermediate latent space is important for achieving high OOD detection accuracy. Our proposed framework, incorporating early stopping, significantly enhances OOD detection efficiency without compromising accuracy, making it practical for real-time applications.<\/jats:p><\/jats:sec>","DOI":"10.3389\/fdata.2024.1444634","type":"journal-article","created":{"date-parts":[[2024,11,20]],"date-time":"2024-11-20T06:32:17Z","timestamp":1732084337000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":2,"title":["Efficient out-of-distribution detection via layer-adaptive scoring and early stopping"],"prefix":"10.3389","volume":"7","author":[{"given":"Haoliang","family":"Wang","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Chen","family":"Zhao","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Feng","family":"Chen","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2024,11,20]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1910.10307","article-title":"Detecting out-of-distribution inputs in deep neural networks using an early-layer output","author":"Abdelzad","year":"2019","journal-title":"arXiv:1910.10307"},{"key":"B2","first-page":"527","article-title":"\u201cAdaptive neural networks for efficient inference,\u201d","volume-title":"ICML","author":"Bolukbasi","year":"2017"},{"key":"B3","first-page":"3606","article-title":"\u201cDescribing textures in the wild,\u201d","volume-title":"CVPR","author":"Cimpoi","year":"2014"},{"key":"B4","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1610.02136","article-title":"A baseline for detecting misclassified and out-of-distribution examples in neural networks","author":"Hendrycks","year":"2016","journal-title":"arXiv:1610.02136"},{"key":"B5","doi-asserted-by":"publisher","first-page":"9844","DOI":"10.48550\/arXiv.1703.09844","article-title":"Multi-scale dense convolutional networks for efficient prediction","volume":"2","author":"Huang","year":"2017","journal-title":"arXiv:1703.09844"},{"key":"B6","first-page":"3301","article-title":"\u201cShallow-deep networks: understanding and mitigating network overthinking,\u201d","volume-title":"ICML","author":"Kaya","year":"2019"},{"key":"B7","first-page":"104","article-title":"\u201cMulti-layer aggregation as a key to feature-based OOD detection,\u201d","volume-title":"International Workshop on Uncertainty for Safe Utilization of Machine Learning in Medical Imaging","author":"Lambert","year":"2023"},{"key":"B8","first-page":"1362","article-title":"\u201cMulti-class data description for out-of-distribution detection,\u201d","volume-title":"KDD","author":"Lee","year":"2020"},{"key":"B9","first-page":"7167","article-title":"\u201cA simple unified framework for detecting out-of-distribution samples and adversarial attacks,\u201d","volume-title":"Adv. Neural Inf. Process. Syst.","author":"Lee","year":"2018"},{"key":"B10","doi-asserted-by":"publisher","first-page":"791","DOI":"10.1007\/s10115-017-1029-1","article-title":"The cascading neural network: building the internet of smart things","volume":"52","author":"Leroux","year":"2017","journal-title":"Knowl. Inform. Syst"},{"key":"B11","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1706.02690","article-title":"Enhancing the reliability of out-of-distribution image detection in neural networks","author":"Liang","year":"2017","journal-title":"arXiv:1706.02690"},{"key":"B12","doi-asserted-by":"crossref","first-page":"413","DOI":"10.1109\/ICDM.2008.17","article-title":"\u201cIsolation forest,\u201d","volume-title":"2008 Eighth IEEE International Conference on Data Mining","author":"Liu","year":"2008"},{"key":"B13","doi-asserted-by":"crossref","first-page":"1049","DOI":"10.1214\/aoms\/1177700079","article-title":"A nonparametric estimate of a multivariate density function","volume":"36","author":"Loftsgaarden","year":"1965","journal-title":"Ann. Math. Stat"},{"key":"B14","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1802.10501","article-title":"Predictive uncertainty estimation via prior networks","author":"Malinin","year":"2018","journal-title":"CoRR, abs\/1802.10501"},{"key":"B15","first-page":"10051","article-title":"\u201cOn the impact of spurious correlation for out-of-distribution detection,\u201d","volume-title":"Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 36","author":"Ming","year":"2022"},{"key":"B16","first-page":"24384","article-title":"\u201cDeep deterministic uncertainty: a new simple baseline,\u201d","volume-title":"Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition","author":"Mukhoti","year":"2023"},{"key":"B17","first-page":"1530","article-title":"\u201cVariational inference with normalizing flows,\u201d","volume-title":"International Conference on Machine Learning","author":"Rezende","year":"2015"},{"key":"B18","doi-asserted-by":"crossref","first-page":"4","DOI":"10.1145\/2689746.2689747","article-title":"\u201cAnomaly detection using autoencoders with nonlinear dimensionality reduction,\u201d","volume-title":"Proceedings of the MLSDA 2014 2nd Workshop on Machine Learning for Sensory Data Analysis","author":"Sakurada","year":"2014"},{"key":"B19","first-page":"582","article-title":"\u201cSupport vector method for novelty detection,\u201d","volume-title":"Adv. Neural Inf. Process. Syst.","author":"Sch\u00f6lkopf","year":"2000"},{"key":"B20","doi-asserted-by":"crossref","DOI":"10.1201\/9781315140919","volume-title":"Density Estimation for Statistics and Data Analysis","author":"Silverman","year":"2018"},{"key":"B21","first-page":"550","article-title":"\u201cOut-of-distribution detection using an ensemble of self supervised leave-out classifiers,\u201d","volume-title":"ECCV","author":"Vyas","year":"2018"},{"key":"B22","doi-asserted-by":"crossref","first-page":"526","DOI":"10.1007\/978-3-031-05936-0_41","article-title":"\u201cLayer adaptive deep neural networks for out-of-distribution detection,\u201d","volume-title":"Pacific-Asia Conference on Knowledge Discovery and Data Mining","author":"Wang","year":"2022"},{"key":"B23","doi-asserted-by":"publisher","first-page":"198","DOI":"10.1016\/j.patcog.2017.09.012","article-title":"Hyperparameter selection of one-class support vector machine by self-adaptive data shifting","volume":"74","author":"Wang","year":"2018","journal-title":"Pat. Recogn"},{"key":"B24","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1504.06755","article-title":"TurkerGaze: crowdsourcing saliency with webcam based eye tracking","author":"Xu","year":"2015","journal-title":"arXiv:1504.06755"},{"key":"B25","doi-asserted-by":"crossref","DOI":"10.1109\/ICDM.2019.00102","article-title":"\u201cRank-based multi-task learning for fair regression,\u201d","volume-title":"IEEE International Conference on Data Mining (ICDM)","author":"Zhao","year":"2019"},{"key":"B26","doi-asserted-by":"publisher","DOI":"10.1145\/3447548.3467389","article-title":"Fairness-aware online meta-learning","author":"Zhao","year":"2021","journal-title":"ACM SIGKDD"},{"key":"B27","doi-asserted-by":"publisher","first-page":"2131","DOI":"10.48550\/arXiv.1711.05611","article-title":"Interpreting deep visual representations via network dissection","volume":"41","author":"Zhou","year":"2018","journal-title":"PAMI"}],"container-title":["Frontiers in Big Data"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fdata.2024.1444634\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2024,11,20]],"date-time":"2024-11-20T06:32:25Z","timestamp":1732084345000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fdata.2024.1444634\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2024,11,20]]},"references-count":27,"alternative-id":["10.3389\/fdata.2024.1444634"],"URL":"https:\/\/doi.org\/10.3389\/fdata.2024.1444634","relation":{},"ISSN":["2624-909X"],"issn-type":[{"value":"2624-909X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2024,11,20]]},"article-number":"1444634"}}