{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:WDHWFHIPHVO2HFFHRGYFARYG7K","short_pith_number":"pith:WDHWFHIP","schema_version":"1.0","canonical_sha256":"b0cf629d0f3d5da394a789b0504706fab4d634607b83fbd04f49d7c9d896c3d5","source":{"kind":"arxiv","id":"2506.03189","version":1},"attestation_state":"computed","paper":{"title":"Continual Learning in Vision-Language Models via Aligned Model Merging","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Ahmet Iscen, Anurag Arnab, Cordelia Schmid, Ghada Sokar, Gintare Karolina Dziugaite, Pablo Samuel Castro","submitted_at":"2025-05-30T20:52:21Z","abstract_excerpt":"Continual learning is conventionally tackled through sequential fine-tuning, a process that, while enabling adaptation, inherently favors plasticity over the stability needed to retain prior knowledge. While existing approaches attempt to mitigate catastrophic forgetting, a bias towards recent tasks persists as they build upon this sequential nature. In this work we present a new perspective based on model merging to maintain stability while still retaining plasticity. Rather than just sequentially updating the model weights, we propose merging newly trained task parameters with previously lea"},"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":"2506.03189","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-05-30T20:52:21Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"f45433956da9d80a2f1e269954d8ba8160e19b1f64e4c588eec9c003130107c8","abstract_canon_sha256":"9537de89ce71974337f082c3ea3fa8375299525efac8f32fb6852a6c5c2603f0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:15:48.113846Z","signature_b64":"A9EVRQyLh/ni5URCHtkL/IkIBEp5/sIJ03rZaZqzyZkmPjfxjCvyhd1bW21PmGnVwQ7Ev5sbCNQBjzjbj63aDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b0cf629d0f3d5da394a789b0504706fab4d634607b83fbd04f49d7c9d896c3d5","last_reissued_at":"2026-07-05T11:15:48.113331Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:15:48.113331Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Continual Learning in Vision-Language Models via Aligned Model Merging","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Ahmet Iscen, Anurag Arnab, Cordelia Schmid, Ghada Sokar, Gintare Karolina Dziugaite, Pablo Samuel Castro","submitted_at":"2025-05-30T20:52:21Z","abstract_excerpt":"Continual learning is conventionally tackled through sequential fine-tuning, a process that, while enabling adaptation, inherently favors plasticity over the stability needed to retain prior knowledge. While existing approaches attempt to mitigate catastrophic forgetting, a bias towards recent tasks persists as they build upon this sequential nature. In this work we present a new perspective based on model merging to maintain stability while still retaining plasticity. Rather than just sequentially updating the model weights, we propose merging newly trained task parameters with previously lea"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.03189","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/2506.03189/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":"2506.03189","created_at":"2026-07-05T11:15:48.113391+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.03189v1","created_at":"2026-07-05T11:15:48.113391+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.03189","created_at":"2026-07-05T11:15:48.113391+00:00"},{"alias_kind":"pith_short_12","alias_value":"WDHWFHIPHVO2","created_at":"2026-07-05T11:15:48.113391+00:00"},{"alias_kind":"pith_short_16","alias_value":"WDHWFHIPHVO2HFFH","created_at":"2026-07-05T11:15:48.113391+00:00"},{"alias_kind":"pith_short_8","alias_value":"WDHWFHIP","created_at":"2026-07-05T11:15:48.113391+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.02214","citing_title":"Unlocking Speech-Text Compositional Powers: Instruction-Following Speech Language Models without Instruction Tuning","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12655","citing_title":"Amnesia: A Stealthy Replay Attack on Continual Learning Dreams","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12419","citing_title":"ORBIT: Preserving Foundational Language Capabilities in GenRetrieval via Origin-Regulated Merging","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12935","citing_title":"Task Alignment: A Simple Proxy for Practical Model Merging Across Diverse Vision Tasks","ref_index":46,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WDHWFHIPHVO2HFFHRGYFARYG7K","json":"https://pith.science/pith/WDHWFHIPHVO2HFFHRGYFARYG7K.json","graph_json":"https://pith.science/api/pith-number/WDHWFHIPHVO2HFFHRGYFARYG7K/graph.json","events_json":"https://pith.science/api/pith-number/WDHWFHIPHVO2HFFHRGYFARYG7K/events.json","paper":"https://pith.science/paper/WDHWFHIP"},"agent_actions":{"view_html":"https://pith.science/pith/WDHWFHIPHVO2HFFHRGYFARYG7K","download_json":"https://pith.science/pith/WDHWFHIPHVO2HFFHRGYFARYG7K.json","view_paper":"https://pith.science/paper/WDHWFHIP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.03189&json=true","fetch_graph":"https://pith.science/api/pith-number/WDHWFHIPHVO2HFFHRGYFARYG7K/graph.json","fetch_events":"https://pith.science/api/pith-number/WDHWFHIPHVO2HFFHRGYFARYG7K/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WDHWFHIPHVO2HFFHRGYFARYG7K/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WDHWFHIPHVO2HFFHRGYFARYG7K/action/storage_attestation","attest_author":"https://pith.science/pith/WDHWFHIPHVO2HFFHRGYFARYG7K/action/author_attestation","sign_citation":"https://pith.science/pith/WDHWFHIPHVO2HFFHRGYFARYG7K/action/citation_signature","submit_replication":"https://pith.science/pith/WDHWFHIPHVO2HFFHRGYFARYG7K/action/replication_record"}},"created_at":"2026-07-05T11:15:48.113391+00:00","updated_at":"2026-07-05T11:15:48.113391+00:00"}