{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DNT325EH5RXT7FCQB42EYV2MX4","short_pith_number":"pith:DNT325EH","schema_version":"1.0","canonical_sha256":"1b67bd7487ec6f3f94500f344c574cbf149b6e1d768a5e594a06637c4ab24457","source":{"kind":"arxiv","id":"2504.10957","version":3},"attestation_state":"computed","paper":{"title":"When is Task Vector Provably Effective for Model Editing? A Generalization Analysis of Nonlinear Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hongkang Li, Meng Wang, Pin-Yu Chen, Shuai Zhang, Sijia Liu, Yihua Zhang","submitted_at":"2025-04-15T08:04:39Z","abstract_excerpt":"Task arithmetic refers to editing the pre-trained model by adding a weighted sum of task vectors, each of which is the weight update from the pre-trained model to fine-tuned models for certain tasks. This approach recently gained attention as a computationally efficient inference method for model editing, e.g., multi-task learning, forgetting, and out-of-domain generalization capabilities. However, the theoretical understanding of why task vectors can execute various conceptual operations remains limited, due to the highly non-convexity of training Transformer-based models. To the best of our "},"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":"2504.10957","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-04-15T08:04:39Z","cross_cats_sorted":[],"title_canon_sha256":"c676b1e9e54d86bcdd24f44ae5f3801bc6a2a5ef6d11b7261d312a59483b47fa","abstract_canon_sha256":"e95406206b98f75dc7282881f024cc7cfe168e55f815fd6540c365decfe3522d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:09:11.452156Z","signature_b64":"9UeFMQyhDNtO39I64P+au/Kc5x1coidEk0RUP3fpYGBrJrnSER7NncbQt+6abwgdKMzB957WoV+oWN7JSbnsAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1b67bd7487ec6f3f94500f344c574cbf149b6e1d768a5e594a06637c4ab24457","last_reissued_at":"2026-07-05T11:09:11.451646Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:09:11.451646Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When is Task Vector Provably Effective for Model Editing? A Generalization Analysis of Nonlinear Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hongkang Li, Meng Wang, Pin-Yu Chen, Shuai Zhang, Sijia Liu, Yihua Zhang","submitted_at":"2025-04-15T08:04:39Z","abstract_excerpt":"Task arithmetic refers to editing the pre-trained model by adding a weighted sum of task vectors, each of which is the weight update from the pre-trained model to fine-tuned models for certain tasks. This approach recently gained attention as a computationally efficient inference method for model editing, e.g., multi-task learning, forgetting, and out-of-domain generalization capabilities. However, the theoretical understanding of why task vectors can execute various conceptual operations remains limited, due to the highly non-convexity of training Transformer-based models. To the best of our "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.10957","kind":"arxiv","version":3},"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/2504.10957/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":"2504.10957","created_at":"2026-07-05T11:09:11.451707+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.10957v3","created_at":"2026-07-05T11:09:11.451707+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.10957","created_at":"2026-07-05T11:09:11.451707+00:00"},{"alias_kind":"pith_short_12","alias_value":"DNT325EH5RXT","created_at":"2026-07-05T11:09:11.451707+00:00"},{"alias_kind":"pith_short_16","alias_value":"DNT325EH5RXT7FCQ","created_at":"2026-07-05T11:09:11.451707+00:00"},{"alias_kind":"pith_short_8","alias_value":"DNT325EH","created_at":"2026-07-05T11:09:11.451707+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.20730","citing_title":"Distributional Alignment as a Criterion for Designing Task Vectors in In-Context Learning","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05710","citing_title":"On the Blessing of Pre-training in Weak-to-Strong Generalization","ref_index":135,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DNT325EH5RXT7FCQB42EYV2MX4","json":"https://pith.science/pith/DNT325EH5RXT7FCQB42EYV2MX4.json","graph_json":"https://pith.science/api/pith-number/DNT325EH5RXT7FCQB42EYV2MX4/graph.json","events_json":"https://pith.science/api/pith-number/DNT325EH5RXT7FCQB42EYV2MX4/events.json","paper":"https://pith.science/paper/DNT325EH"},"agent_actions":{"view_html":"https://pith.science/pith/DNT325EH5RXT7FCQB42EYV2MX4","download_json":"https://pith.science/pith/DNT325EH5RXT7FCQB42EYV2MX4.json","view_paper":"https://pith.science/paper/DNT325EH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.10957&json=true","fetch_graph":"https://pith.science/api/pith-number/DNT325EH5RXT7FCQB42EYV2MX4/graph.json","fetch_events":"https://pith.science/api/pith-number/DNT325EH5RXT7FCQB42EYV2MX4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DNT325EH5RXT7FCQB42EYV2MX4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DNT325EH5RXT7FCQB42EYV2MX4/action/storage_attestation","attest_author":"https://pith.science/pith/DNT325EH5RXT7FCQB42EYV2MX4/action/author_attestation","sign_citation":"https://pith.science/pith/DNT325EH5RXT7FCQB42EYV2MX4/action/citation_signature","submit_replication":"https://pith.science/pith/DNT325EH5RXT7FCQB42EYV2MX4/action/replication_record"}},"created_at":"2026-07-05T11:09:11.451707+00:00","updated_at":"2026-07-05T11:09:11.451707+00:00"}