{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:UM3GFLDZHHTV6ONWGZK7ZX4EXH","short_pith_number":"pith:UM3GFLDZ","schema_version":"1.0","canonical_sha256":"a33662ac7939e75f39b63655fcdf84b9cf79ad896dd6aca956ebbc00c5155163","source":{"kind":"arxiv","id":"2103.02079","version":2},"attestation_state":"computed","paper":{"title":"DP-InstaHide: Provably Defusing Poisoning and Backdoor Attacks with Differentially Private Data Augmentations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR","cs.CV"],"primary_cat":"cs.LG","authors_text":"Amin Ghiasi, Arjun Gupta, Eitan Borgnia, Furong Huang, Jonas Geiping, Liam Fowl, Micah Goldblum, Tom Goldstein, Valeriia Cherepanova","submitted_at":"2021-03-02T23:07:31Z","abstract_excerpt":"Data poisoning and backdoor attacks manipulate training data to induce security breaches in a victim model. These attacks can be provably deflected using differentially private (DP) training methods, although this comes with a sharp decrease in model performance. The InstaHide method has recently been proposed as an alternative to DP training that leverages supposed privacy properties of the mixup augmentation, although without rigorous guarantees. In this work, we show that strong data augmentations, such as mixup and random additive noise, nullify poison attacks while enduring only a small a"},"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":"2103.02079","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-03-02T23:07:31Z","cross_cats_sorted":["cs.CR","cs.CV"],"title_canon_sha256":"5f7b99ab3556d5bceda3847421d67cdfdff0b0a641b46e54f8c2def3bbc05363","abstract_canon_sha256":"0f9abc10b22e246b2cb57f6356502f87f1a5b1d05e73512cf45db56c090550f6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:30:15.194684Z","signature_b64":"KAJZh24qp4aE7i+qMbZ7LL2a4LQ8VsAlYIInSdpl8tNwlZ938OpvBvd3df90J/QYCQmyZtLC93tumr1XuGGiAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a33662ac7939e75f39b63655fcdf84b9cf79ad896dd6aca956ebbc00c5155163","last_reissued_at":"2026-07-05T02:30:15.194173Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:30:15.194173Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DP-InstaHide: Provably Defusing Poisoning and Backdoor Attacks with Differentially Private Data Augmentations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR","cs.CV"],"primary_cat":"cs.LG","authors_text":"Amin Ghiasi, Arjun Gupta, Eitan Borgnia, Furong Huang, Jonas Geiping, Liam Fowl, Micah Goldblum, Tom Goldstein, Valeriia Cherepanova","submitted_at":"2021-03-02T23:07:31Z","abstract_excerpt":"Data poisoning and backdoor attacks manipulate training data to induce security breaches in a victim model. These attacks can be provably deflected using differentially private (DP) training methods, although this comes with a sharp decrease in model performance. The InstaHide method has recently been proposed as an alternative to DP training that leverages supposed privacy properties of the mixup augmentation, although without rigorous guarantees. In this work, we show that strong data augmentations, such as mixup and random additive noise, nullify poison attacks while enduring only a small a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2103.02079","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/2103.02079/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":"2103.02079","created_at":"2026-07-05T02:30:15.194229+00:00"},{"alias_kind":"arxiv_version","alias_value":"2103.02079v2","created_at":"2026-07-05T02:30:15.194229+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2103.02079","created_at":"2026-07-05T02:30:15.194229+00:00"},{"alias_kind":"pith_short_12","alias_value":"UM3GFLDZHHTV","created_at":"2026-07-05T02:30:15.194229+00:00"},{"alias_kind":"pith_short_16","alias_value":"UM3GFLDZHHTV6ONW","created_at":"2026-07-05T02:30:15.194229+00:00"},{"alias_kind":"pith_short_8","alias_value":"UM3GFLDZ","created_at":"2026-07-05T02:30:15.194229+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.17373","citing_title":"FML-bench: A Controlled Study of AI Research Agent Strategies from the Perspective of Search Dynamics","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17373","citing_title":"FML-bench: A Controlled Study of AI Research Agent Strategies from the Perspective of Search Dynamics","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UM3GFLDZHHTV6ONWGZK7ZX4EXH","json":"https://pith.science/pith/UM3GFLDZHHTV6ONWGZK7ZX4EXH.json","graph_json":"https://pith.science/api/pith-number/UM3GFLDZHHTV6ONWGZK7ZX4EXH/graph.json","events_json":"https://pith.science/api/pith-number/UM3GFLDZHHTV6ONWGZK7ZX4EXH/events.json","paper":"https://pith.science/paper/UM3GFLDZ"},"agent_actions":{"view_html":"https://pith.science/pith/UM3GFLDZHHTV6ONWGZK7ZX4EXH","download_json":"https://pith.science/pith/UM3GFLDZHHTV6ONWGZK7ZX4EXH.json","view_paper":"https://pith.science/paper/UM3GFLDZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2103.02079&json=true","fetch_graph":"https://pith.science/api/pith-number/UM3GFLDZHHTV6ONWGZK7ZX4EXH/graph.json","fetch_events":"https://pith.science/api/pith-number/UM3GFLDZHHTV6ONWGZK7ZX4EXH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UM3GFLDZHHTV6ONWGZK7ZX4EXH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UM3GFLDZHHTV6ONWGZK7ZX4EXH/action/storage_attestation","attest_author":"https://pith.science/pith/UM3GFLDZHHTV6ONWGZK7ZX4EXH/action/author_attestation","sign_citation":"https://pith.science/pith/UM3GFLDZHHTV6ONWGZK7ZX4EXH/action/citation_signature","submit_replication":"https://pith.science/pith/UM3GFLDZHHTV6ONWGZK7ZX4EXH/action/replication_record"}},"created_at":"2026-07-05T02:30:15.194229+00:00","updated_at":"2026-07-05T02:30:15.194229+00:00"}