Composing predictable primitives for zero-shot learning

Composing predictable primitives for zero-shot learning | 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 Composing predictable primitives for zero-shot learning Pablo Tano, Jacob Bakermans, Charles Findling, Tiago Branco, Alexandre Pouget This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8879475/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Animals exhibit a remarkable capacity to adapt to novel challenges on their first attempt. We propose that this zero-shot adaptability requires keeping behavioral outcomes highly predictable and controllable, which is achieved by structuring behavior as short compositions of simple primitives. Continuously across environments, the agent learns to predict the outcomes of primitive sequences in an unsupervised manner. When solving a new task, it optimizes short primitive compositions using differentiable model-predictive control. Constraining behavior to short compositions keeps gradient-based planning tractable, yielding efficiency orders of magnitude beyond traditional methods. Applied to mouse homing-escape behavior, our model explains the emergence of subgoals across arena configurations where conventional models fail. Ablations confirm that the components enabling rapid learning in our model (world models, planning models, primitives, and compositionality) are necessary to reproduce escape trajectories. Our results suggest that optimizing predictable primitive compositions is a core mechanism driving zero-shot adaptability and explaining mouse behavioral phenomena. Biological sciences/Neuroscience/Computational neuroscience/Learning algorithms Biological sciences/Neuroscience/Motor control Full Text Additional Declarations There is NO Competing Interest. Cite Share Download PDF Status: Under Review 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-8879475","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":600134161,"identity":"5a58476b-4e54-4036-bf7b-1cb84d7321fd","order_by":0,"name":"Pablo Tano","email":"data:image/png;base64,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","orcid":"https://orcid.org/0000-0002-5665-8050","institution":"Champalimaud Centre for the Unknown","correspondingAuthor":true,"prefix":"","firstName":"Pablo","middleName":"","lastName":"Tano","suffix":""},{"id":600134162,"identity":"1428c1c8-5768-433b-bf34-89b1fc1d6097","order_by":1,"name":"Jacob Bakermans","email":"","orcid":"https://orcid.org/0000-0003-1645-2645","institution":"University College London","correspondingAuthor":false,"prefix":"","firstName":"Jacob","middleName":"","lastName":"Bakermans","suffix":""},{"id":600134163,"identity":"d66818c5-743e-4a2a-ac15-463d82484a08","order_by":2,"name":"Charles Findling","email":"","orcid":"https://orcid.org/0009-0000-4274-7127","institution":"École Normale Supérieure","correspondingAuthor":false,"prefix":"","firstName":"Charles","middleName":"","lastName":"Findling","suffix":""},{"id":600134164,"identity":"ea99b3a3-e9ee-4b94-bd64-e65cb571710a","order_by":3,"name":"Tiago Branco","email":"","orcid":"https://orcid.org/0000-0001-5087-3465","institution":"UCL Sainsbury Wellcome Centre for Neural Circuits and Behaviour","correspondingAuthor":false,"prefix":"","firstName":"Tiago","middleName":"","lastName":"Branco","suffix":""},{"id":600134165,"identity":"409c826c-e57d-41db-870b-8b7434ee4787","order_by":4,"name":"Alexandre Pouget","email":"","orcid":"https://orcid.org/0000-0003-3054-6365","institution":"University of Geneva","correspondingAuthor":false,"prefix":"","firstName":"Alexandre","middleName":"","lastName":"Pouget","suffix":""}],"badges":[],"createdAt":"2026-02-14 11:20:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8879475/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8879475/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104401585,"identity":"c1fe86b1-6911-43a8-89fe-878939f5d0ef","added_by":"auto","created_at":"2026-03-11 12:13:03","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2451757,"visible":true,"origin":"","legend":"Article File","description":"","filename":"CompPredPrims.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8879475/v1_covered_8c42f410-f131-4ef7-99a1-39597344e169.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Composing predictable primitives for zero-shot learning","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"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":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8879475/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8879475/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Animals exhibit a remarkable capacity to adapt to novel challenges on their first attempt. We propose that this zero-shot adaptability requires keeping behavioral outcomes highly predictable and controllable, which is achieved by structuring behavior as short compositions of simple primitives. Continuously across environments, the agent learns to predict the outcomes of primitive sequences in an unsupervised manner. When solving a new task, it optimizes short primitive compositions using differentiable model-predictive control. Constraining behavior to short compositions keeps gradient-based planning tractable, yielding efficiency orders of magnitude beyond traditional methods. Applied to mouse homing-escape behavior, our model explains the emergence of subgoals across arena configurations where conventional models fail. Ablations confirm that the components enabling rapid learning in our model (world models, planning models, primitives, and compositionality) are necessary to reproduce escape trajectories. Our results suggest that optimizing predictable primitive compositions is a core mechanism driving zero-shot adaptability and explaining mouse behavioral phenomena.","manuscriptTitle":"Composing predictable primitives for zero-shot learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-04 08:02:47","doi":"10.21203/rs.3.rs-8879475/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-neuroscience","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"neuro","sideBox":"Learn more about [Nature Neuroscience](http://www.nature.com/neuro/)","snPcode":"","submissionUrl":"","title":"Nature Neuroscience","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Research","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"d9a13afc-f4ee-48a4-9b66-bd3b408c2d14","owner":[],"postedDate":"March 4th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":63862623,"name":"Biological sciences/Neuroscience/Computational neuroscience/Learning algorithms"},{"id":63862624,"name":"Biological sciences/Neuroscience/Motor control"}],"tags":[],"updatedAt":"2026-04-08T17:51:17+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-04 08:02:47","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8879475","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8879475","identity":"rs-8879475","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}