{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:VDK6MBX55VGO2B46SXUS5ESXZY","short_pith_number":"pith:VDK6MBX5","schema_version":"1.0","canonical_sha256":"a8d5e606fded4ced079e95e92e9257ce3c77e92056f481849826a2b782aa5116","source":{"kind":"arxiv","id":"2605.16834","version":1},"attestation_state":"computed","paper":{"title":"Learning Relative Representations for Fine-Grained Multimodal Alignment with Limited Data","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Relative representations via learnable anchors align token-level structures across modalities using only limited paired examples.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Shiwon Kim, Yu Rang Park","submitted_at":"2026-05-16T06:33:38Z","abstract_excerpt":"Multimodal pre-training demonstrates strong generalization performance, but this paradigm is often impractical in domains where paired data are scarce. A promising alternative is post-hoc multimodal alignment, which aligns separately pre-trained unimodal encoders using a limited number of paired examples. However, existing methods focus primarily on aligning global representations, missing patch-token relations. This may hinder transfer to tasks that require fine-grained cross-modal matching beyond coarse sample-level semantics. To address this issue, we propose a post-hoc alignment method tha"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.16834","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-16T06:33:38Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"e824f1a709c8e7b834f17d7af1268258aae06bbd1521fd7f212830335d200e83","abstract_canon_sha256":"bc1667a62d044b06d87094c993236f37b2668814313f48ae7ffeead1c276eec0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:25.230505Z","signature_b64":"CKhUrcI3D8YC4YuhVAjyNg3iLboRhZTCuaeU6IwAV7SoulbYFQda7SZnRFXTzJt4g7dls1h2S+8V4kbkOJ0yCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a8d5e606fded4ced079e95e92e9257ce3c77e92056f481849826a2b782aa5116","last_reissued_at":"2026-05-20T00:03:25.229675Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:25.229675Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning Relative Representations for Fine-Grained Multimodal Alignment with Limited Data","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Relative representations via learnable anchors align token-level structures across modalities using only limited paired examples.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Shiwon Kim, Yu Rang Park","submitted_at":"2026-05-16T06:33:38Z","abstract_excerpt":"Multimodal pre-training demonstrates strong generalization performance, but this paradigm is often impractical in domains where paired data are scarce. A promising alternative is post-hoc multimodal alignment, which aligns separately pre-trained unimodal encoders using a limited number of paired examples. However, existing methods focus primarily on aligning global representations, missing patch-token relations. This may hinder transfer to tasks that require fine-grained cross-modal matching beyond coarse sample-level semantics. To address this issue, we propose a post-hoc alignment method tha"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Despite learning only the anchors without heavy projection layers, our approach consistently outperforms existing methods in zero-shot classification, cross-modal retrieval, and zero-shot segmentation by a substantial margin.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That training a set of learnable anchors to induce consistent cross-modal similarity patterns for matched pairs is sufficient to capture fine-grained token-level relations without requiring additional projection layers or larger paired datasets.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A new post-hoc alignment technique uses learnable anchors to capture token-level relative similarities between modalities, outperforming global alignment baselines on zero-shot classification, retrieval, and segmentation with scarce paired examples.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Relative representations via learnable anchors align token-level structures across modalities using only limited paired examples.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"777d2eef5a35ef6d232a757a47b54f8b3170a4f641b64bcc37565a76bdddecca"},"source":{"id":"2605.16834","kind":"arxiv","version":1},"verdict":{"id":"6033a144-e12e-4919-a6e6-7670c8ff92b3","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:45:06.663156Z","strongest_claim":"Despite learning only the anchors without heavy projection layers, our approach consistently outperforms existing methods in zero-shot classification, cross-modal retrieval, and zero-shot segmentation by a substantial margin.","one_line_summary":"A new post-hoc alignment technique uses learnable anchors to capture token-level relative similarities between modalities, outperforming global alignment baselines on zero-shot classification, retrieval, and segmentation with scarce paired examples.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That training a set of learnable anchors to induce consistent cross-modal similarity patterns for matched pairs is sufficient to capture fine-grained token-level relations without requiring additional projection layers or larger paired datasets.","pith_extraction_headline":"Relative representations via learnable anchors align token-level structures across modalities using only limited paired examples."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16834/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.248521Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T20:51:12.437387Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T19:01:56.254703Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.398949Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"e3c8db28b098ec01440f520c3b6850965316f6b9e479d244961270d94a7a134f"},"references":{"count":31,"sample":[{"doi":"","year":2021,"title":"Learning transferable visual models from natural language supervision","work_id":"9ff70d47-d644-4c21-a8f2-2e01a0734b97","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Le, Yun- Hsuan Sung, Zhen Li, and Tom Duerig","work_id":"8d222499-da6d-44c6-80d6-d4490667ce96","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"The platonic representation hypothesis","work_id":"7e23235a-add0-43c4-8e54-bb2b253439c8","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Relative representations enable zero-shot latent space communication","work_id":"1ce4feb0-b01e-4d77-93f4-86427be53c00","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Linearly mapping from image to text space.arXiv preprint arXiv:2209.15162","work_id":"1470dc2c-e9da-4e3f-aa62-808e9aa7bc2e","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":31,"snapshot_sha256":"08c7405dc4cdc8623b7973f217df8c480ff121cf8c44e3cdffa9195f18d58927","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"ddf4b06482223e22f26d8fa6ef40834b6630c1fc4b8b0375c5854928715e4f0d"},"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":"2605.16834","created_at":"2026-05-20T00:03:25.229809+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.16834v1","created_at":"2026-05-20T00:03:25.229809+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.16834","created_at":"2026-05-20T00:03:25.229809+00:00"},{"alias_kind":"pith_short_12","alias_value":"VDK6MBX55VGO","created_at":"2026-05-20T00:03:25.229809+00:00"},{"alias_kind":"pith_short_16","alias_value":"VDK6MBX55VGO2B46","created_at":"2026-05-20T00:03:25.229809+00:00"},{"alias_kind":"pith_short_8","alias_value":"VDK6MBX5","created_at":"2026-05-20T00:03:25.229809+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VDK6MBX55VGO2B46SXUS5ESXZY","json":"https://pith.science/pith/VDK6MBX55VGO2B46SXUS5ESXZY.json","graph_json":"https://pith.science/api/pith-number/VDK6MBX55VGO2B46SXUS5ESXZY/graph.json","events_json":"https://pith.science/api/pith-number/VDK6MBX55VGO2B46SXUS5ESXZY/events.json","paper":"https://pith.science/paper/VDK6MBX5"},"agent_actions":{"view_html":"https://pith.science/pith/VDK6MBX55VGO2B46SXUS5ESXZY","download_json":"https://pith.science/pith/VDK6MBX55VGO2B46SXUS5ESXZY.json","view_paper":"https://pith.science/paper/VDK6MBX5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.16834&json=true","fetch_graph":"https://pith.science/api/pith-number/VDK6MBX55VGO2B46SXUS5ESXZY/graph.json","fetch_events":"https://pith.science/api/pith-number/VDK6MBX55VGO2B46SXUS5ESXZY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VDK6MBX55VGO2B46SXUS5ESXZY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VDK6MBX55VGO2B46SXUS5ESXZY/action/storage_attestation","attest_author":"https://pith.science/pith/VDK6MBX55VGO2B46SXUS5ESXZY/action/author_attestation","sign_citation":"https://pith.science/pith/VDK6MBX55VGO2B46SXUS5ESXZY/action/citation_signature","submit_replication":"https://pith.science/pith/VDK6MBX55VGO2B46SXUS5ESXZY/action/replication_record"}},"created_at":"2026-05-20T00:03:25.229809+00:00","updated_at":"2026-05-20T00:03:25.229809+00:00"}