{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Z4AKNNPHLKAAAWPYAQ7D5N2LNI","short_pith_number":"pith:Z4AKNNPH","schema_version":"1.0","canonical_sha256":"cf00a6b5e75a800059f8043e3eb74b6a10d8e7bd41898e43ca61255b4fdc787a","source":{"kind":"arxiv","id":"2507.15836","version":1},"attestation_state":"computed","paper":{"title":"Optimizing Canaries for Privacy Auditing with Metagradient Descent","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Andrew Ilyas, Matteo Boglioni, Terrance Liu, Zhiwei Steven Wu","submitted_at":"2025-07-21T17:47:33Z","abstract_excerpt":"In this work we study black-box privacy auditing, where the goal is to lower bound the privacy parameter of a differentially private learning algorithm using only the algorithm's outputs (i.e., final trained model). For DP-SGD (the most successful method for training differentially private deep learning models), the canonical approach auditing uses membership inference-an auditor comes with a small set of special \"canary\" examples, inserts a random subset of them into the training set, and then tries to discern which of their canaries were included in the training set (typically via a membersh"},"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":"2507.15836","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-07-21T17:47:33Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"325b066aa69358f9d44995eb7d3c9fd93faa2671e5a2c58b5acae68fe6d3403e","abstract_canon_sha256":"801af3c3b8e2ad04e5d268fd9ac472312f7c8738951efafac345b76aca54bea9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:40:45.515674Z","signature_b64":"qqYQCTF4ZRwRP5d5s1ieqoyhvnzyKtnbUKteWu7ipptD6sYC6YFz9+7FGnB3YtWPqhbyWOWRiICu57urHi79AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cf00a6b5e75a800059f8043e3eb74b6a10d8e7bd41898e43ca61255b4fdc787a","last_reissued_at":"2026-07-05T11:40:45.515135Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:40:45.515135Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Optimizing Canaries for Privacy Auditing with Metagradient Descent","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Andrew Ilyas, Matteo Boglioni, Terrance Liu, Zhiwei Steven Wu","submitted_at":"2025-07-21T17:47:33Z","abstract_excerpt":"In this work we study black-box privacy auditing, where the goal is to lower bound the privacy parameter of a differentially private learning algorithm using only the algorithm's outputs (i.e., final trained model). For DP-SGD (the most successful method for training differentially private deep learning models), the canonical approach auditing uses membership inference-an auditor comes with a small set of special \"canary\" examples, inserts a random subset of them into the training set, and then tries to discern which of their canaries were included in the training set (typically via a membersh"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.15836","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/2507.15836/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":"2507.15836","created_at":"2026-07-05T11:40:45.515194+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.15836v1","created_at":"2026-07-05T11:40:45.515194+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.15836","created_at":"2026-07-05T11:40:45.515194+00:00"},{"alias_kind":"pith_short_12","alias_value":"Z4AKNNPHLKAA","created_at":"2026-07-05T11:40:45.515194+00:00"},{"alias_kind":"pith_short_16","alias_value":"Z4AKNNPHLKAAAWPY","created_at":"2026-07-05T11:40:45.515194+00:00"},{"alias_kind":"pith_short_8","alias_value":"Z4AKNNPH","created_at":"2026-07-05T11:40:45.515194+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/Z4AKNNPHLKAAAWPYAQ7D5N2LNI","json":"https://pith.science/pith/Z4AKNNPHLKAAAWPYAQ7D5N2LNI.json","graph_json":"https://pith.science/api/pith-number/Z4AKNNPHLKAAAWPYAQ7D5N2LNI/graph.json","events_json":"https://pith.science/api/pith-number/Z4AKNNPHLKAAAWPYAQ7D5N2LNI/events.json","paper":"https://pith.science/paper/Z4AKNNPH"},"agent_actions":{"view_html":"https://pith.science/pith/Z4AKNNPHLKAAAWPYAQ7D5N2LNI","download_json":"https://pith.science/pith/Z4AKNNPHLKAAAWPYAQ7D5N2LNI.json","view_paper":"https://pith.science/paper/Z4AKNNPH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.15836&json=true","fetch_graph":"https://pith.science/api/pith-number/Z4AKNNPHLKAAAWPYAQ7D5N2LNI/graph.json","fetch_events":"https://pith.science/api/pith-number/Z4AKNNPHLKAAAWPYAQ7D5N2LNI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Z4AKNNPHLKAAAWPYAQ7D5N2LNI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Z4AKNNPHLKAAAWPYAQ7D5N2LNI/action/storage_attestation","attest_author":"https://pith.science/pith/Z4AKNNPHLKAAAWPYAQ7D5N2LNI/action/author_attestation","sign_citation":"https://pith.science/pith/Z4AKNNPHLKAAAWPYAQ7D5N2LNI/action/citation_signature","submit_replication":"https://pith.science/pith/Z4AKNNPHLKAAAWPYAQ7D5N2LNI/action/replication_record"}},"created_at":"2026-07-05T11:40:45.515194+00:00","updated_at":"2026-07-05T11:40:45.515194+00:00"}