{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:CYH5EEAM2EMGRCQF2LG3GE3AFX","short_pith_number":"pith:CYH5EEAM","schema_version":"1.0","canonical_sha256":"160fd2100cd118688a05d2cdb313602ddfb6eb33d2cd8abe3fd644ef99bedf09","source":{"kind":"arxiv","id":"2206.02770","version":1},"attestation_state":"computed","paper":{"title":"Multimodal Contrastive Learning with LIMoE: the Language-Image Mixture of Experts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Basil Mustafa, Carlos Riquelme, Joan Puigcerver, Neil Houlsby, Rodolphe Jenatton","submitted_at":"2022-06-06T17:51:59Z","abstract_excerpt":"Large sparsely-activated models have obtained excellent performance in multiple domains. However, such models are typically trained on a single modality at a time. We present the Language-Image MoE, LIMoE, a sparse mixture of experts model capable of multimodal learning. LIMoE accepts both images and text simultaneously, while being trained using a contrastive loss. MoEs are a natural fit for a multimodal backbone, since expert layers can learn an appropriate partitioning of modalities. However, new challenges arise; in particular, training stability and balanced expert utilization, for which "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2206.02770","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2022-06-06T17:51:59Z","cross_cats_sorted":[],"title_canon_sha256":"92511804ddf7e20c26665451f6ad04f8261626d70e13203bc7f2bbaa0bbbac5a","abstract_canon_sha256":"f223261f50ade1d7ae3ed80d406b127c43fa13b12a433403ee9299a275589f7e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:29:21.070435Z","signature_b64":"1vyGxrtTmnY9Fyw0z/Nashff8W9LuNfNlMI+VoV8/jsPRftc/yjEpuQASGOrf6mDVvU8mzZPCAAXpRTyGEbdBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"160fd2100cd118688a05d2cdb313602ddfb6eb33d2cd8abe3fd644ef99bedf09","last_reissued_at":"2026-07-05T04:29:21.070082Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:29:21.070082Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multimodal Contrastive Learning with LIMoE: the Language-Image Mixture of Experts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Basil Mustafa, Carlos Riquelme, Joan Puigcerver, Neil Houlsby, Rodolphe Jenatton","submitted_at":"2022-06-06T17:51:59Z","abstract_excerpt":"Large sparsely-activated models have obtained excellent performance in multiple domains. However, such models are typically trained on a single modality at a time. We present the Language-Image MoE, LIMoE, a sparse mixture of experts model capable of multimodal learning. LIMoE accepts both images and text simultaneously, while being trained using a contrastive loss. MoEs are a natural fit for a multimodal backbone, since expert layers can learn an appropriate partitioning of modalities. However, new challenges arise; in particular, training stability and balanced expert utilization, for which "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2206.02770","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2206.02770/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2206.02770","created_at":"2026-07-05T04:29:21.070141+00:00"},{"alias_kind":"arxiv_version","alias_value":"2206.02770v1","created_at":"2026-07-05T04:29:21.070141+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2206.02770","created_at":"2026-07-05T04:29:21.070141+00:00"},{"alias_kind":"pith_short_12","alias_value":"CYH5EEAM2EMG","created_at":"2026-07-05T04:29:21.070141+00:00"},{"alias_kind":"pith_short_16","alias_value":"CYH5EEAM2EMGRCQF","created_at":"2026-07-05T04:29:21.070141+00:00"},{"alias_kind":"pith_short_8","alias_value":"CYH5EEAM","created_at":"2026-07-05T04:29:21.070141+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24985","citing_title":"Retrieval-Augmented Personalization with Foundation Models for Wearable Stress Detection","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20610","citing_title":"Beyond Routing: Characterising Expert Tuning and Representation in Vision Mixture-of-Experts","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CYH5EEAM2EMGRCQF2LG3GE3AFX","json":"https://pith.science/pith/CYH5EEAM2EMGRCQF2LG3GE3AFX.json","graph_json":"https://pith.science/api/pith-number/CYH5EEAM2EMGRCQF2LG3GE3AFX/graph.json","events_json":"https://pith.science/api/pith-number/CYH5EEAM2EMGRCQF2LG3GE3AFX/events.json","paper":"https://pith.science/paper/CYH5EEAM"},"agent_actions":{"view_html":"https://pith.science/pith/CYH5EEAM2EMGRCQF2LG3GE3AFX","download_json":"https://pith.science/pith/CYH5EEAM2EMGRCQF2LG3GE3AFX.json","view_paper":"https://pith.science/paper/CYH5EEAM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2206.02770&json=true","fetch_graph":"https://pith.science/api/pith-number/CYH5EEAM2EMGRCQF2LG3GE3AFX/graph.json","fetch_events":"https://pith.science/api/pith-number/CYH5EEAM2EMGRCQF2LG3GE3AFX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CYH5EEAM2EMGRCQF2LG3GE3AFX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CYH5EEAM2EMGRCQF2LG3GE3AFX/action/storage_attestation","attest_author":"https://pith.science/pith/CYH5EEAM2EMGRCQF2LG3GE3AFX/action/author_attestation","sign_citation":"https://pith.science/pith/CYH5EEAM2EMGRCQF2LG3GE3AFX/action/citation_signature","submit_replication":"https://pith.science/pith/CYH5EEAM2EMGRCQF2LG3GE3AFX/action/replication_record"}},"created_at":"2026-07-05T04:29:21.070141+00:00","updated_at":"2026-07-05T04:29:21.070141+00:00"}