{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,22]],"date-time":"2026-07-22T06:06:27Z","timestamp":1784700387563,"version":"3.55.0"},"reference-count":43,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,7,22]],"date-time":"2026-07-22T00:00:00Z","timestamp":1784678400000},"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. Artif. Intell."],"abstract":"<jats:sec>\n                    <jats:title>Introduction<\/jats:title>\n                    <jats:p>Neural Architecture Search (NAS) effectively automates Deep Learning pipeline design but is prone to validation overfitting when applied to complex tasks, such as medical image analysis. To mitigate this and enhance generalization, researchers frequently integrate Deep Ensemble Learning (DEL) and data augmentation into the NAS workflow. However, the assumption that these methodologies do not negatively interfere in high-overfitting scenarios remains unproven.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>We evaluated NAS, combined with DEL and data augmentation pipelines, across both CIFAR-10 and a biomedical CT dataset, assessing the influence of task complexity and data scarcity in their interaction. Using an ablative experimental design, we isolated the contributions of DEL and NAS and mapped data augmentation sensitivity landscapes at both global and local scales.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>While synergistic on large, well defined datasets, statistical analysis revealed that DEL failed to significantly enhance the generalization capabilities of NAS-generated populations in data-constrained regimes. Moreover, we identified local roughness within the data augmentation sensitivity landscapes.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Discussion<\/jats:title>\n                    <jats:p>Our findings challenge the prevailing assumption of unconditional methodological synergy that guides joint architecture exploration, ensemble pruning and data augmentation optimization.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/frai.2026.1869820","type":"journal-article","created":{"date-parts":[[2026,7,22]],"date-time":"2026-07-22T05:52:31Z","timestamp":1784699551000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["On the fragility of neural architecture search: the role of overfitting and task complexity in medical image analysis"],"prefix":"10.3389","volume":"9","author":[{"given":"A.","family":"G\u00f3mez","sequence":"first","affiliation":[{"name":"Department of Bioengineering, Universidad Carlos III de Madrid","place":["Madrid, Spain"]},{"name":"Unidad de Medicina y Cirug\u00eda Experimental, Instituto de Investigaci\u00f3n Sanitaria Gregorio Mara\u00f1\u00f3n, Hospital General Universitario Gregorio Mara\u00f1\u00f3n","place":["Madrid, Spain"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"M.","family":"Desco","sequence":"additional","affiliation":[{"name":"Department of Bioengineering, Universidad Carlos III de Madrid","place":["Madrid, Spain"]},{"name":"Unidad de Medicina y Cirug\u00eda Experimental, Instituto de Investigaci\u00f3n Sanitaria Gregorio Mara\u00f1\u00f3n, Hospital General Universitario Gregorio Mara\u00f1\u00f3n","place":["Madrid, Spain"]},{"name":"Centro de Investigaci\u00f3n Biom\u00e9dica en Red de Salud Mental (CIBERSAM), Instituto de Salud Carlos III","place":["Madrid, Spain"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"M.","family":"Abella","sequence":"additional","affiliation":[{"name":"Department of Bioengineering, Universidad Carlos III de Madrid","place":["Madrid, Spain"]},{"name":"Unidad de Medicina y Cirug\u00eda Experimental, Instituto de Investigaci\u00f3n Sanitaria Gregorio Mara\u00f1\u00f3n, Hospital General Universitario Gregorio Mara\u00f1\u00f3n","place":["Madrid, Spain"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,7,22]]},"reference":[{"key":"ref1","doi-asserted-by":"publisher","first-page":"6007","DOI":"10.1002\/mp.15065","article-title":"Tolerance to geometrical inaccuracies in CBCT systems: a comprehensive study","volume":"48","author":"Abella","year":"2021","journal-title":"Med. Phys."},{"key":"ref2","doi-asserted-by":"publisher","first-page":"e0180363","DOI":"10.1371\/journal.pone.0180363","article-title":"FUX-Sim: implementation of a fast universal simulation\/reconstruction framework for X-ray systems","volume":"12","author":"Abella","year":"2017","journal-title":"PLoS One"},{"key":"ref3","article-title":"Bayesian optimized deep ensemble for uncertainty quantification of deep neural networks: a system safety case study on sodium fast reactor thermal stratification modeling","volume-title":"arXiv","author":"Abulawi","year":"2024"},{"key":"ref4","doi-asserted-by":"publisher","first-page":"e715","DOI":"10.7717\/peerj-cs.715","article-title":"Deepening into the suitability of using pre-trained models of ImageNet against a lightweight convolutional neural network in medical imaging: an experimental study","volume":"7","author":"Alzubaidi","year":"2021","journal-title":"PeerJ Comput. Sci."},{"key":"ref5","doi-asserted-by":"crossref","DOI":"10.52202\/068431-2745","article-title":"The effects of regularization and data augmentation are class dependent","volume-title":"arXiv","author":"Balestriero","year":"2022"},{"key":"ref6","doi-asserted-by":"publisher","first-page":"103","DOI":"10.3233\/his-2007-4204","article-title":"On diversity and accuracy of homogeneous and heterogeneous ensembles","volume":"4","author":"Bian","year":"2007","journal-title":"Int. J. Hybrid Intell. Syst."},{"key":"ref7","volume-title":"Bias, Variance, and Arcing Classifiers. Technical Report 460","author":"Breiman","year":"2000"},{"key":"ref8","first-page":"2079","article-title":"On over-fitting in model selection and subsequent selection bias in performance evaluation","volume":"11","author":"Cawley","year":"2010","journal-title":"J. Mach. Learn. Res."},{"key":"ref9","first-page":"16525","author":"Chen","year":"2021"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2010.01412","article-title":"Sharpness-aware minimization for efficiently improving generalization","author":"Foret","year":"2021","journal-title":"arXiv"},{"key":"ref11","doi-asserted-by":"publisher","first-page":"105151","DOI":"10.1016\/j.engappai.2022.105151","article-title":"Ensemble deep learning: a review","volume":"115","author":"Ganaie","year":"2022","journal-title":"Eng. Appl. Artif. Intell."},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2011.11778","article-title":"KeepAugment: a simple information-preserving data augmentation approach","author":"Gong","year":"2020","journal-title":"arXiv"},{"key":"ref13","first-page":"52","author":"Guerra-Librero","year":"2023"},{"key":"ref14","doi-asserted-by":"publisher","first-page":"237","DOI":"10.1016\/j.neucom.2017.06.052","article-title":"Margin & diversity based ordering ensemble pruning","volume":"275","author":"Guo","year":"2017","journal-title":"Neurocomputing"},{"key":"ref15","first-page":"344","author":"He","year":"2023"},{"key":"ref16","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1909.09148","article-title":"Data augmentation revisited: rethinking the distribution gap between clean and augmented data","author":"He","year":"2019","journal-title":"arXiv"},{"key":"ref17","author":"Hendrycks","year":"2018"},{"key":"ref18","article-title":"Data augmentation instead of explicit regularization","volume-title":"arXiv","author":"Hern\u00e1ndez-Garc\u00eda","year":"2018"},{"key":"ref19","volume-title":"High Performance Computing","author":"Herron","year":"2020"},{"key":"ref20","volume-title":"Breakthroughs in Statistics: Methodology and Distribution","author":"Huber","year":"1992"},{"key":"ref21","doi-asserted-by":"publisher","first-page":"794","DOI":"10.1016\/j.jksuci.2018.09.017","article-title":"Artificial bee colony algorithm: a component-wise analysis using diversity measurement","volume":"32","author":"Hussain","year":"2020","journal-title":"J. King Saud Univ. -Comput. Inf. Sci."},{"key":"ref22","doi-asserted-by":"publisher","first-page":"130288","DOI":"10.1016\/j.neucom.2025.130288","article-title":"Sample-customized implicit semantic data augmentation for neural networks regularization","volume":"640","author":"Kang","year":"2025","journal-title":"Neurocomputing"},{"key":"ref23","doi-asserted-by":"publisher","first-page":"459","DOI":"10.1007\/s10898-007-9149-x","article-title":"A powerful and efficient algorithm for numerical function optimization: artificial bee colony (ABC) algorithm","volume":"39","author":"Karaboga","year":"2007","journal-title":"J. Glob. Optim."},{"key":"ref24","article-title":"Joint search of data augmentation policies and network architectures","volume-title":"arXiv","author":"Kashima","year":"2020"},{"key":"ref25","first-page":"15","author":"Kishor","year":"2015"},{"key":"ref26","volume-title":"Learning Multiple Layers of Features from Tiny Images","author":"Krizhevsky","year":"2009"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1807.00263","article-title":"Accurate uncertainties for deep learning using calibrated regression","author":"Kuleshov","year":"2018","journal-title":"arXiv"},{"key":"ref28","doi-asserted-by":"publisher","first-page":"181","DOI":"10.1023\/A:1022859003006","article-title":"Measures of diversity in classifier ensembles and their relationship with the ensemble accuracy","volume":"51","author":"Kuncheva","year":"2003","journal-title":"Mach. Learn."},{"key":"ref29","doi-asserted-by":"publisher","first-page":"2278","DOI":"10.1109\/5.726791","article-title":"Gradient-based learning applied to document recognition","volume":"86","author":"Lecun","year":"1998","journal-title":"Proc. IEEE"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1712.09913","article-title":"Visualizing the loss landscape of neural nets","author":"Li","year":"2018","journal-title":"arXiv"},{"key":"ref31","doi-asserted-by":"publisher","first-page":"2561","DOI":"10.3390\/biomedicines12112561","article-title":"Transfer learning approaches for brain metastases screenings","volume":"12","author":"Luu","year":"2024","journal-title":"Biomedicine"},{"key":"ref32","first-page":"211","author":"Margineantu","year":"1997"},{"key":"ref33","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2512.04992","article-title":"Evolutionary architecture search through grammar-based sequence alignment","author":"Mart\u00edn","year":"2025","journal-title":"arXiv"},{"key":"ref34","volume-title":"Aggregation Ordering in Bagging","author":"Mart\u00ednez-Mu\u00f1oz","year":"2004"},{"key":"ref35","doi-asserted-by":"publisher","first-page":"100258","DOI":"10.1016\/j.array.2022.100258","article-title":"Data augmentation: a comprehensive survey of modern approaches","volume":"16","author":"Mumuni","year":"2022","journal-title":"Array"},{"key":"ref36","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2008.11687","article-title":"What is being transferred in transfer learning?","author":"Neyshabur","year":"2021","journal-title":"arXiv"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2210.14016","article-title":"Shortest edit path crossover: a theory-driven solution to the permutation problem in evolutionary neural architecture search","author":"Qiu","year":"2023","journal-title":"arXiv"},{"key":"ref38","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1902.07208","article-title":"Transfusion: understanding transfer learning for medical imaging","author":"Raghu","year":"2019","journal-title":"arXiv"},{"key":"ref39","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1409.1556","article-title":"Very deep convolutional networks for large-scale image recognition","author":"Simonyan","year":"2014","journal-title":"arXiv"},{"key":"ref40","first-page":"464","author":"Smith","year":"2017"},{"key":"ref41","doi-asserted-by":"publisher","first-page":"127664","DOI":"10.1016\/j.neucom.2024.127664","article-title":"Multi-population evolutionary neural architecture search with stacked generalization","volume":"587","author":"Song","year":"2024","journal-title":"Neurocomputing"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2301.08727","article-title":"Neural architecture search: insights from 1000 papers","author":"White","year":"2023","journal-title":"arXiv"},{"key":"ref43","first-page":"7898","author":"Zaidi","year":"2021"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1869820\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,7,22]],"date-time":"2026-07-22T05:52:35Z","timestamp":1784699555000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1869820\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,7,22]]},"references-count":43,"alternative-id":["10.3389\/frai.2026.1869820"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1869820","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,7,22]]},"article-number":"1869820"}}