Sparse Full-Dimensional Self-attention: Used for Long-Term Human Posture Prediction

Sparse Full-Dimensional Self-attention: Used for Long-Term Human Posture Prediction | 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 Research Article Sparse Full-Dimensional Self-attention: Used for Long-Term Human Posture Prediction Xianhua Li, zhen liu, ShuoYu Tian This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5687102/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Researchers have conducted extensive investigations into transformers, which exhibit strong performance in time series prediction problems. Nevertheless, existing methodologies still encounter challenges when it comes to capturing long-range dependencies. In this study, we introduce a novel sparse full-dimensional attention mechanism known as the "Amformer" to address this issue and enable accurate long-term human movement prediction. The core of the Amformer lies in establishing relationships between feature vectors at the current time and those at different time points, facilitating the efficient fusion of spatiotemporal features. This approach excels at capturing dependencies over extended distances, resulting in more precise predictive outcomes. We propose an uncorrelated initialization strategy for the network to reduce convolutional ambiguity concerning time features effectively. Furthermore, we introduce a time enhancement method to prevent prediction results from converging to a specific intermediate attitude. This method proves beneficial in mitigating the issue of the network gravitating towards an intermediate stance during the prediction process, thereby enhancing the stability and accuracy of prediction results. Our proposed approach outperformed benchmark methods in long-term prediction tasks, as validated through experiments on the Human3.6M dataset. Cross spatiotemporal feature fusion network Human motion prediction Attention mechanism Long time series prediction Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 23 Dec, 2024 Reviewers invited by journal 23 Dec, 2024 Editor assigned by journal 23 Dec, 2024 Submission checks completed at journal 23 Dec, 2024 First submitted to journal 20 Dec, 2024 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-5687102","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":393892450,"identity":"2ebabe80-eaf7-4f2a-af13-bec4dbbef2de","order_by":0,"name":"Xianhua Li","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAqElEQVRIie3QsQrCMBCA4YZAp0jWBnyI20Qo+iopBScHH+FKXyKCz+F8IXtdhb5EujtY6eIiXDeH++f7OO6KQpL+sUojeaiNtcgmCilfTlsXaAVRIaca0DMFjF2XDDwMFKTydGYQd4s4k9HsNGp3vTOIrZqF7JFKveGQciGDAfJM8tkSA9AK4kIzPxla40LsebfAs03Zvw5Ha/uYJw75SuG6eUmSJOl3bzg6OSToAJIPAAAAAElFTkSuQmCC","orcid":"","institution":"First Affiliated Hospital of Anhui University of Science and Technology (Huainan First People's Hospital)","correspondingAuthor":true,"prefix":"","firstName":"Xianhua","middleName":"","lastName":"Li","suffix":""},{"id":393892452,"identity":"28c44342-9ce8-4cae-bb96-354bb4339381","order_by":1,"name":"zhen liu","email":"","orcid":"","institution":"School of Artificial Intelligence, Anhui University of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"zhen","middleName":"","lastName":"liu","suffix":""},{"id":393892453,"identity":"a07a4627-7bbb-4af9-95ab-e967bf5fbaa7","order_by":2,"name":"ShuoYu Tian","email":"","orcid":"","institution":"School of Mechanical and Electrical Engineering, Anhui University of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"ShuoYu","middleName":"","lastName":"Tian","suffix":""}],"badges":[],"createdAt":"2024-12-21 00:23:07","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5687102/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5687102/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":72324038,"identity":"4d44e7b2-655a-45df-9c28-23252a837106","added_by":"auto","created_at":"2024-12-25 11:14:13","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":578824,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5687102/v1_covered_ca9e2391-79e3-4870-8b22-7465adb7961c.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Sparse Full-Dimensional Self-attention: Used for Long-Term Human Posture Prediction","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"signal-image-and-video-processing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"sivp","sideBox":"Learn more about [Signal, Image and Video Processing](http://link.springer.com/journal/11760)","snPcode":"11760","submissionUrl":"https://submission.nature.com/new-submission/11760/3","title":"Signal, Image and Video Processing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Cross spatiotemporal feature fusion network, Human motion prediction, Attention mechanism, Long time series prediction","lastPublishedDoi":"10.21203/rs.3.rs-5687102/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5687102/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eResearchers have conducted extensive investigations into transformers, which exhibit strong performance in time series prediction problems. Nevertheless, existing methodologies still encounter challenges when it comes to capturing long-range dependencies. In this study, we introduce a novel sparse full-dimensional attention mechanism known as the \"Amformer\" to address this issue and enable accurate long-term human movement prediction. The core of the Amformer lies in establishing relationships between feature vectors at the current time and those at different time points, facilitating the efficient fusion of spatiotemporal features. This approach excels at capturing dependencies over extended distances, resulting in more precise predictive outcomes. We propose an uncorrelated initialization strategy for the network to reduce convolutional ambiguity concerning time features effectively. Furthermore, we introduce a time enhancement method to prevent prediction results from converging to a specific intermediate attitude. This method proves beneficial in mitigating the issue of the network gravitating towards an intermediate stance during the prediction process, thereby enhancing the stability and accuracy of prediction results. Our proposed approach outperformed benchmark methods in long-term prediction tasks, as validated through experiments on the Human3.6M dataset.\u003c/p\u003e","manuscriptTitle":"Sparse Full-Dimensional Self-attention: Used for Long-Term Human Posture Prediction","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-25 10:58:08","doi":"10.21203/rs.3.rs-5687102/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-12-23T21:20:25+00:00","index":"","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-12-23T21:19:39+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-12-23T17:30:07+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-12-23T17:28:43+00:00","index":"","fulltext":""},{"type":"submitted","content":"Signal, Image and Video Processing","date":"2024-12-21T00:12:50+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"signal-image-and-video-processing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"sivp","sideBox":"Learn more about [Signal, Image and Video Processing](http://link.springer.com/journal/11760)","snPcode":"11760","submissionUrl":"https://submission.nature.com/new-submission/11760/3","title":"Signal, Image and Video Processing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"c1160294-2ac3-4a30-a164-3037c764c65a","owner":[],"postedDate":"December 25th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-09-26T19:53:10+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-25 10:58:08","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5687102","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5687102","identity":"rs-5687102","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}