Hybrid Neural Tangent Kernel–SGD Optimization for Robust and Scalable Deep Learning Across Medical, Sensor, and Image Domains

Hybrid Neural Tangent Kernel–SGD Optimization for Robust and Scalable Deep Learning Across Medical, Sensor, and Image Domains | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Hybrid Neural Tangent Kernel–SGD Optimization for Robust and Scalable Deep Learning Across Medical, Sensor, and Image Domains Ahmed Mubaraki This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7566608/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Efficient and reliable training of deep neural networks remains a major challenge, particularly in fields such as medical imaging, sensor analysis, and ecological monitoring where data are often noisy, high-dimensional, and heterogeneous. Here we introduce a hybrid optimization framework that integrates Neural Tangent Kernel (NTK) theory with Stochastic Gradient Descent (SGD). The approach leverages NTK-based initialization to stabilize early training, followed by SGD fine-tuning to enable adaptability and scalability. I evaluate this method on diverse datasets including ISIC 2018 (skin lesion analysis), NIH ChestX-ray14 (radiography), Tiny ImageNet (natural image classification), and UCI HAR (sensorbased activity recognition). Across all benchmarks, the hybrid NTK-SGD method consistently outperforms NTK-only baselines and matches or exceeds standard SGD, while delivering faster convergence and improved robustness to adversarial noise. By uniting theory-driven stability with data-driven flexibility, NTKSGD offers a generalizable, interpretable, and computationally efficient training strategy. These results highlight its potential for cross-domain deployment in medical, environmental, and industrial AI applications, where both accuracy and resilience are critical. Biological sciences/Computational biology and bioinformatics Physical sciences/Engineering Physical sciences/Mathematics and computing Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7566608","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":567538428,"identity":"d3d06749-2b71-4846-be02-ccdd89526ff6","order_by":0,"name":"Ahmed Mubaraki","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2UlEQVRIiWNgGAWjYHACNhBhwMbeAKKZSdHCc4CxgTQtDBIJRGrhn3bG7MHPPXbGfJJvzB8wVFgnNvAffoBXi8TtHHPDnmfJZmzSOYYNDGfSExsk0gzwW3M7x0yC5wCzDVgLY9thoBYG/FrkgVok/xyot2GTPAPU8g+ohf/4B7xaDIBapHkOHDZjk+ABamkAamHIwW+L4e20cmOZA8eN2XjSCmckHEs3bpPIKcCrRe528raHbw5UG85vP7zhw4caa9l+/uMb8GpBBQkM0GgaBaNgFIyCUUAZAADmw0LfThhS5gAAAABJRU5ErkJggg==","orcid":"","institution":"University of Tabuk","correspondingAuthor":true,"prefix":"","firstName":"Ahmed","middleName":"","lastName":"Mubaraki","suffix":""}],"badges":[],"createdAt":"2025-09-08 17:23:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7566608/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7566608/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":99326484,"identity":"794eea2a-2ca8-45cc-abe4-abe8084e29ad","added_by":"auto","created_at":"2025-12-31 16:50:54","extension":"json","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4051,"visible":true,"origin":"","legend":"","description":"","filename":"477ebe43cbfe4729a017946c446cc5d9.json","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/2a609a30826b08573e4651ba.json"},{"id":99326406,"identity":"a71132bc-c077-453d-a663-1b146795ae44","added_by":"auto","created_at":"2025-12-31 16:50:47","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":165211,"visible":true,"origin":"","legend":"","description":"","filename":"PointbyPointResponse.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/e5070f15ee9c16f3f3439b8e.pdf"},{"id":99326381,"identity":"d278e0b8-3b0c-4572-9276-070e9dc4907b","added_by":"auto","created_at":"2025-12-31 16:50:46","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2254696,"visible":true,"origin":"","legend":"","description":"","filename":"TemplateforsubmissionstoScientificReports.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/e9b06960ca081cb048c00a47.pdf"},{"id":99326438,"identity":"aea5c608-aaf4-48a2-ac6a-67b69489f101","added_by":"auto","created_at":"2025-12-31 16:50:48","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":105381,"visible":true,"origin":"","legend":"","description":"","filename":"ablationcomparison.png","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/f32df62bd5a06fb67976e6db.png"},{"id":99326509,"identity":"125976dc-045a-4112-b0c9-2c56034d24c0","added_by":"auto","created_at":"2025-12-31 16:50:58","extension":"ldf","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":49107,"visible":true,"origin":"","legend":"","description":"","filename":"jabbrvltwaall.ldf","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/e2fe48a5afa15288b91d892f.ldf"},{"id":99326435,"identity":"4868388e-e6b5-4dab-b587-87f342c31a91","added_by":"auto","created_at":"2025-12-31 16:50:48","extension":"ldf","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":268657,"visible":true,"origin":"","legend":"","description":"","filename":"jabbrvltwaen.ldf","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/d88d2b20d80573e44e00bf92.ldf"},{"id":99326378,"identity":"4da839c4-d445-48af-8e5b-6e80b804c548","added_by":"auto","created_at":"2025-12-31 16:50:45","extension":"sty","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2500,"visible":true,"origin":"","legend":"","description":"","filename":"jabbrv.sty","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/30c0f5db22ca0dbf6ca77cdc.sty"},{"id":99326407,"identity":"3182f14a-0028-48e0-a4ff-b7b6b172ce1c","added_by":"auto","created_at":"2025-12-31 16:50:47","extension":"bst","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":36153,"visible":true,"origin":"","legend":"","description":"","filename":"naturemagdoi.bst","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/c583fe1f602a465a3b6c33dd.bst"},{"id":99326480,"identity":"cff5f5d4-190a-4a01-9894-870fc5db5bad","added_by":"auto","created_at":"2025-12-31 16:50:53","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":56740,"visible":true,"origin":"","legend":"","description":"","filename":"ntktrainingpipelinestylized1.png","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/5768de70f5b12604fe261859.png"},{"id":99326343,"identity":"e7697691-1b37-4dcd-a188-d5643ea71329","added_by":"auto","created_at":"2025-12-31 16:50:42","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1340280,"visible":true,"origin":"","legend":"","description":"","filename":"ntktrainingworkflowacademic1.png","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/879609470694c3832e542304.png"},{"id":99326377,"identity":"572939f5-48b2-48f0-98e8-52fec2dd44a6","added_by":"auto","created_at":"2025-12-31 16:50:45","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":123184,"visible":true,"origin":"","legend":"","description":"","filename":"resourceefficiencycomparison.png","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/6c01f8f063c330343f38ece2.png"},{"id":99326375,"identity":"aef0e3e1-0f61-4f2a-ad36-ff30da9c07c0","added_by":"auto","created_at":"2025-12-31 16:50:45","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":180653,"visible":true,"origin":"","legend":"","description":"","filename":"updatedrobustnesscomparison.png","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/2ee717dfc55c07b07b44707d.png"},{"id":99326405,"identity":"382feab7-bd97-49bd-868c-74fef0936786","added_by":"auto","created_at":"2025-12-31 16:50:47","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":89879,"visible":true,"origin":"","legend":"","description":"","filename":"updatedtrainingtimecomparison.png","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/dd578314bb03d987db20ac83.png"},{"id":99326445,"identity":"5df0b332-88ee-4a38-a7b3-4597aed04e3d","added_by":"auto","created_at":"2025-12-31 16:50:50","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":139536,"visible":true,"origin":"","legend":"","description":"","filename":"updatedtsnecomparison.png","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/8c8e424b9dcc1406a8e84794.png"},{"id":99326320,"identity":"1442b41f-347a-4ae6-8372-9e35bf547804","added_by":"auto","created_at":"2025-12-31 16:50:41","extension":"cls","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5824,"visible":true,"origin":"","legend":"","description":"","filename":"wlscirep.cls","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/8f9af7b76b1f5c5d8a159b73.cls"},{"id":99326436,"identity":"21393a5e-98d4-4624-8965-75823406f566","added_by":"auto","created_at":"2025-12-31 16:50:48","extension":"xml","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":52094,"visible":true,"origin":"","legend":"","description":"","filename":"477ebe43cbfe4729a017946c446cc5d91structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1/232387a5019beb8c58efaad5.xml"},{"id":99790235,"identity":"ad3c78c3-51d1-4698-aff8-ab0748ac3ef3","added_by":"auto","created_at":"2026-01-08 12:57:27","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":993768,"visible":true,"origin":"","legend":"","description":"","filename":"TemplateforsubmissionstoScientificReports.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7566608/v1_covered_feda862f-2f3e-472f-ab50-989af10485c5.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Hybrid Neural Tangent Kernel–SGD Optimization for Robust and Scalable Deep Learning Across Medical, Sensor, and Image Domains","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7566608/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7566608/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eEfficient and reliable training of deep neural networks remains a major challenge, particularly in fields such as medical imaging, sensor analysis, and ecological monitoring where data are often noisy, high-dimensional, and heterogeneous. Here we introduce a hybrid optimization framework that integrates Neural Tangent Kernel (NTK) theory with Stochastic Gradient Descent (SGD). The approach leverages NTK-based initialization to stabilize early training, followed by SGD fine-tuning to enable adaptability and scalability.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eI evaluate this method on diverse datasets including ISIC 2018 (skin lesion analysis), NIH ChestX-ray14 (radiography), Tiny ImageNet (natural image classification), and UCI HAR (sensorbased activity recognition). Across all benchmarks, the hybrid NTK-SGD method consistently outperforms NTK-only baselines and matches or exceeds standard SGD, while delivering faster convergence and improved robustness to adversarial noise.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBy uniting theory-driven stability with data-driven flexibility, NTKSGD offers a generalizable, interpretable, and computationally efficient training strategy. These results highlight its potential for cross-domain deployment in medical, environmental, and industrial AI applications, where both accuracy and resilience are critical.\u003c/p\u003e","manuscriptTitle":"Hybrid Neural Tangent Kernel–SGD Optimization for Robust and Scalable Deep Learning Across Medical, Sensor, and Image Domains","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-31 16:40:17","doi":"10.21203/rs.3.rs-7566608/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2ec4c6db-e7bc-46f2-922e-a1332c5b702b","owner":[],"postedDate":"December 31st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":60399561,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":60399562,"name":"Physical sciences/Engineering"},{"id":60399563,"name":"Physical sciences/Mathematics and computing"}],"tags":[],"updatedAt":"2026-01-05T06:39:33+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-31 16:40:17","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7566608","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7566608","identity":"rs-7566608","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}