{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:MQLF6WX2M4IZB772EBQYQKLKHJ","short_pith_number":"pith:MQLF6WX2","schema_version":"1.0","canonical_sha256":"64165f5afa671190fffa206188296a3a58931a1450612cc360869e297a5e1756","source":{"kind":"arxiv","id":"2504.16430","version":1},"attestation_state":"computed","paper":{"title":"MAGIC: Near-Optimal Data Attribution for Deep Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Andrew Ilyas, Logan Engstrom","submitted_at":"2025-04-23T05:32:37Z","abstract_excerpt":"The goal of predictive data attribution is to estimate how adding or removing a given set of training datapoints will affect model predictions. In convex settings, this goal is straightforward (i.e., via the infinitesimal jackknife). In large-scale (non-convex) settings, however, existing methods are far less successful -- current methods' estimates often only weakly correlate with ground truth. In this work, we present a new data attribution method (MAGIC) that combines classical methods and recent advances in metadifferentiation to (nearly) optimally estimate the effect of adding or removing"},"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":"2504.16430","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-04-23T05:32:37Z","cross_cats_sorted":["cs.CL","cs.CV","stat.ML"],"title_canon_sha256":"671866c480260a45342dd06960d2d4437761d0a7b602ac96c82f475e358189ea","abstract_canon_sha256":"eb7becc657418229f1772d2b36e7f3babf5e99b2b0a75d940653f43ff798003f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:52:46.569044Z","signature_b64":"bPUeyufIVD6KnkiCgepw5QgAB/TQUJl6onUeu2CWo0PYXfPwCL7U/7b8f81zPkxu4epjX021uNd0w9eayQ0vDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"64165f5afa671190fffa206188296a3a58931a1450612cc360869e297a5e1756","last_reissued_at":"2026-07-05T10:52:46.568531Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:52:46.568531Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MAGIC: Near-Optimal Data Attribution for Deep Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Andrew Ilyas, Logan Engstrom","submitted_at":"2025-04-23T05:32:37Z","abstract_excerpt":"The goal of predictive data attribution is to estimate how adding or removing a given set of training datapoints will affect model predictions. In convex settings, this goal is straightforward (i.e., via the infinitesimal jackknife). In large-scale (non-convex) settings, however, existing methods are far less successful -- current methods' estimates often only weakly correlate with ground truth. In this work, we present a new data attribution method (MAGIC) that combines classical methods and recent advances in metadifferentiation to (nearly) optimally estimate the effect of adding or removing"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.16430","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/2504.16430/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":"2504.16430","created_at":"2026-07-05T10:52:46.568595+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.16430v1","created_at":"2026-07-05T10:52:46.568595+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.16430","created_at":"2026-07-05T10:52:46.568595+00:00"},{"alias_kind":"pith_short_12","alias_value":"MQLF6WX2M4IZ","created_at":"2026-07-05T10:52:46.568595+00:00"},{"alias_kind":"pith_short_16","alias_value":"MQLF6WX2M4IZB772","created_at":"2026-07-05T10:52:46.568595+00:00"},{"alias_kind":"pith_short_8","alias_value":"MQLF6WX2","created_at":"2026-07-05T10:52:46.568595+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.00510","citing_title":"Prototype Language Models","ref_index":163,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06533","citing_title":"Position: Don't Just \"Fix it in Post\": A Science of AI Must Study Training Dynamics","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2606.24890","citing_title":"Small edits, large models: How Wikipedia advocacy shapes LLM values","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18814","citing_title":"How Faithful Is Trajectory-Based Data Attribution? Error Sources, Remedies, and Practical Guidelines","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2602.03783","citing_title":"Efficient Estimation of Kernel Surrogate Models for Task Attribution","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07328","citing_title":"How to sketch a learning algorithm","ref_index":10,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MQLF6WX2M4IZB772EBQYQKLKHJ","json":"https://pith.science/pith/MQLF6WX2M4IZB772EBQYQKLKHJ.json","graph_json":"https://pith.science/api/pith-number/MQLF6WX2M4IZB772EBQYQKLKHJ/graph.json","events_json":"https://pith.science/api/pith-number/MQLF6WX2M4IZB772EBQYQKLKHJ/events.json","paper":"https://pith.science/paper/MQLF6WX2"},"agent_actions":{"view_html":"https://pith.science/pith/MQLF6WX2M4IZB772EBQYQKLKHJ","download_json":"https://pith.science/pith/MQLF6WX2M4IZB772EBQYQKLKHJ.json","view_paper":"https://pith.science/paper/MQLF6WX2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.16430&json=true","fetch_graph":"https://pith.science/api/pith-number/MQLF6WX2M4IZB772EBQYQKLKHJ/graph.json","fetch_events":"https://pith.science/api/pith-number/MQLF6WX2M4IZB772EBQYQKLKHJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MQLF6WX2M4IZB772EBQYQKLKHJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MQLF6WX2M4IZB772EBQYQKLKHJ/action/storage_attestation","attest_author":"https://pith.science/pith/MQLF6WX2M4IZB772EBQYQKLKHJ/action/author_attestation","sign_citation":"https://pith.science/pith/MQLF6WX2M4IZB772EBQYQKLKHJ/action/citation_signature","submit_replication":"https://pith.science/pith/MQLF6WX2M4IZB772EBQYQKLKHJ/action/replication_record"}},"created_at":"2026-07-05T10:52:46.568595+00:00","updated_at":"2026-07-05T10:52:46.568595+00:00"}