{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:PIQUSN3L4LNJRGFM3NRQNARXPQ","short_pith_number":"pith:PIQUSN3L","schema_version":"1.0","canonical_sha256":"7a2149376be2da9898acdb630682377c1ad71d31efc600aabcdfd76aa47967e7","source":{"kind":"arxiv","id":"2602.05536","version":2},"attestation_state":"computed","paper":{"title":"When Shared Knowledge Hurts: Spectral Over-Accumulation in Model Merging","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"Jian Zhang, Jintao Guo, Yayuan Li, Yinghuan Shi, Yue Duan, Ze Peng","submitted_at":"2026-02-05T10:52:36Z","abstract_excerpt":"Model merging combines multiple fine-tuned models into a single model by adding their weight updates, providing a lightweight alternative to retraining. Existing methods primarily target resolving conflicts between task updates, leaving the failure mode of over-counting shared knowledge unaddressed. We show that when tasks share aligned spectral directions (i.e., overlapping singular vectors), a simple linear combination repeatedly accumulates these directions, inflating the singular values and biasing the merged model toward shared subspaces. To mitigate this issue, we propose Singular Value "},"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":"2602.05536","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-05T10:52:36Z","cross_cats_sorted":["cs.AI","cs.CL","cs.CV"],"title_canon_sha256":"ca70526108cce7b1bbfa74085a5f6d867ee456839d9313bead76c34d809e7716","abstract_canon_sha256":"14e987228ee238ecafbf4e8d12941f89fc7a3de7aa13feb76cabd65165a35ae5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-22T01:04:54.856127Z","signature_b64":"rQZWqT4QoPQhzYdKGwpK94/a7UZ2bWGKhaQdI02QgoUwEG7fAqD9tGbgOuJbjgAVVqfHzNQoqD4Fbkx1K6RGAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7a2149376be2da9898acdb630682377c1ad71d31efc600aabcdfd76aa47967e7","last_reissued_at":"2026-05-22T01:04:54.855532Z","signature_status":"signed_v1","first_computed_at":"2026-05-22T01:04:54.855532Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Shared Knowledge Hurts: Spectral Over-Accumulation in Model Merging","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"Jian Zhang, Jintao Guo, Yayuan Li, Yinghuan Shi, Yue Duan, Ze Peng","submitted_at":"2026-02-05T10:52:36Z","abstract_excerpt":"Model merging combines multiple fine-tuned models into a single model by adding their weight updates, providing a lightweight alternative to retraining. Existing methods primarily target resolving conflicts between task updates, leaving the failure mode of over-counting shared knowledge unaddressed. We show that when tasks share aligned spectral directions (i.e., overlapping singular vectors), a simple linear combination repeatedly accumulates these directions, inflating the singular values and biasing the merged model toward shared subspaces. To mitigate this issue, we propose Singular Value "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.05536","kind":"arxiv","version":2},"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/2602.05536/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":"2602.05536","created_at":"2026-05-22T01:04:54.855613+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.05536v2","created_at":"2026-05-22T01:04:54.855613+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.05536","created_at":"2026-05-22T01:04:54.855613+00:00"},{"alias_kind":"pith_short_12","alias_value":"PIQUSN3L4LNJ","created_at":"2026-05-22T01:04:54.855613+00:00"},{"alias_kind":"pith_short_16","alias_value":"PIQUSN3L4LNJRGFM","created_at":"2026-05-22T01:04:54.855613+00:00"},{"alias_kind":"pith_short_8","alias_value":"PIQUSN3L","created_at":"2026-05-22T01:04:54.855613+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.19852","citing_title":"Are Tools Always Beneficial? Learning to Invoke Tools Adaptively for Dual-Mode Multimodal LLM Reasoning","ref_index":92,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PIQUSN3L4LNJRGFM3NRQNARXPQ","json":"https://pith.science/pith/PIQUSN3L4LNJRGFM3NRQNARXPQ.json","graph_json":"https://pith.science/api/pith-number/PIQUSN3L4LNJRGFM3NRQNARXPQ/graph.json","events_json":"https://pith.science/api/pith-number/PIQUSN3L4LNJRGFM3NRQNARXPQ/events.json","paper":"https://pith.science/paper/PIQUSN3L"},"agent_actions":{"view_html":"https://pith.science/pith/PIQUSN3L4LNJRGFM3NRQNARXPQ","download_json":"https://pith.science/pith/PIQUSN3L4LNJRGFM3NRQNARXPQ.json","view_paper":"https://pith.science/paper/PIQUSN3L","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.05536&json=true","fetch_graph":"https://pith.science/api/pith-number/PIQUSN3L4LNJRGFM3NRQNARXPQ/graph.json","fetch_events":"https://pith.science/api/pith-number/PIQUSN3L4LNJRGFM3NRQNARXPQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PIQUSN3L4LNJRGFM3NRQNARXPQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PIQUSN3L4LNJRGFM3NRQNARXPQ/action/storage_attestation","attest_author":"https://pith.science/pith/PIQUSN3L4LNJRGFM3NRQNARXPQ/action/author_attestation","sign_citation":"https://pith.science/pith/PIQUSN3L4LNJRGFM3NRQNARXPQ/action/citation_signature","submit_replication":"https://pith.science/pith/PIQUSN3L4LNJRGFM3NRQNARXPQ/action/replication_record"}},"created_at":"2026-05-22T01:04:54.855613+00:00","updated_at":"2026-05-22T01:04:54.855613+00:00"}