{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:VKL566SMWE7OGMNPOVMNB6SH3S","short_pith_number":"pith:VKL566SM","schema_version":"1.0","canonical_sha256":"aa97df7a4cb13ee331af7558d0fa47dca72264cc4819c00b8e327c0cbec46d13","source":{"kind":"arxiv","id":"2607.08194","version":1},"attestation_state":"computed","paper":{"title":"Dive Into the Implicit Biases of Low-rank Vision-language Alignment","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Anyang Su, Chenxu Zhao, Kai Wang, Minghui Wu, Mingjia Shi, Ping Jiang, Shuo Wang, Sifan Zhou, Tianyu Fu, Xiaobo Wang","submitted_at":"2026-07-09T07:50:49Z","abstract_excerpt":"Vision-language alignment, the stage that bridges pretrained vision encoders and large language models, is widely treated as a form of pretraining requiring full-parameter updates. We challenge this view and investigate what happens when low-rank adaptation is applied to the LLM during this stage instead. We find that low-rank alignment not only reduces computational costs but also outperforms full-parameter alignment on most benchmarks. To understand this phenomenon, we systematically characterize the implicit biases introduced by low-rank adaptation during alignment. Empirically, we find 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":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2607.08194","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CV","submitted_at":"2026-07-09T07:50:49Z","cross_cats_sorted":[],"title_canon_sha256":"3b3bbc4e101bfb089e31a9527c8206cde7e7f868fa822bce00bf78d061383e17","abstract_canon_sha256":"f74b60128db88889b3287a776eff6fd8caa32e379e243fafee8cb5e864697d2b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-10T01:19:29.433886Z","signature_b64":"6e5Uq5/+Qjg1UraRwVsb5Sp+OPp7kHWTo3pmJ69/7elFbfakYkPTE2YAIgM+n738OrX28xUzrCySA5Hw49cyBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"aa97df7a4cb13ee331af7558d0fa47dca72264cc4819c00b8e327c0cbec46d13","last_reissued_at":"2026-07-10T01:19:29.433365Z","signature_status":"signed_v1","first_computed_at":"2026-07-10T01:19:29.433365Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dive Into the Implicit Biases of Low-rank Vision-language Alignment","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Anyang Su, Chenxu Zhao, Kai Wang, Minghui Wu, Mingjia Shi, Ping Jiang, Shuo Wang, Sifan Zhou, Tianyu Fu, Xiaobo Wang","submitted_at":"2026-07-09T07:50:49Z","abstract_excerpt":"Vision-language alignment, the stage that bridges pretrained vision encoders and large language models, is widely treated as a form of pretraining requiring full-parameter updates. We challenge this view and investigate what happens when low-rank adaptation is applied to the LLM during this stage instead. We find that low-rank alignment not only reduces computational costs but also outperforms full-parameter alignment on most benchmarks. To understand this phenomenon, we systematically characterize the implicit biases introduced by low-rank adaptation during alignment. Empirically, we find tha"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.08194","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/2607.08194/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":"2607.08194","created_at":"2026-07-10T01:19:29.433427+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.08194v1","created_at":"2026-07-10T01:19:29.433427+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.08194","created_at":"2026-07-10T01:19:29.433427+00:00"},{"alias_kind":"pith_short_12","alias_value":"VKL566SMWE7O","created_at":"2026-07-10T01:19:29.433427+00:00"},{"alias_kind":"pith_short_16","alias_value":"VKL566SMWE7OGMNP","created_at":"2026-07-10T01:19:29.433427+00:00"},{"alias_kind":"pith_short_8","alias_value":"VKL566SM","created_at":"2026-07-10T01:19:29.433427+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VKL566SMWE7OGMNPOVMNB6SH3S","json":"https://pith.science/pith/VKL566SMWE7OGMNPOVMNB6SH3S.json","graph_json":"https://pith.science/api/pith-number/VKL566SMWE7OGMNPOVMNB6SH3S/graph.json","events_json":"https://pith.science/api/pith-number/VKL566SMWE7OGMNPOVMNB6SH3S/events.json","paper":"https://pith.science/paper/VKL566SM"},"agent_actions":{"view_html":"https://pith.science/pith/VKL566SMWE7OGMNPOVMNB6SH3S","download_json":"https://pith.science/pith/VKL566SMWE7OGMNPOVMNB6SH3S.json","view_paper":"https://pith.science/paper/VKL566SM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.08194&json=true","fetch_graph":"https://pith.science/api/pith-number/VKL566SMWE7OGMNPOVMNB6SH3S/graph.json","fetch_events":"https://pith.science/api/pith-number/VKL566SMWE7OGMNPOVMNB6SH3S/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VKL566SMWE7OGMNPOVMNB6SH3S/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VKL566SMWE7OGMNPOVMNB6SH3S/action/storage_attestation","attest_author":"https://pith.science/pith/VKL566SMWE7OGMNPOVMNB6SH3S/action/author_attestation","sign_citation":"https://pith.science/pith/VKL566SMWE7OGMNPOVMNB6SH3S/action/citation_signature","submit_replication":"https://pith.science/pith/VKL566SMWE7OGMNPOVMNB6SH3S/action/replication_record"}},"created_at":"2026-07-10T01:19:29.433427+00:00","updated_at":"2026-07-10T01:19:29.433427+00:00"}