{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2021:QQXQN67SCKJ36AMMDGKOLPVMAH","short_pith_number":"pith:QQXQN67S","canonical_record":{"source":{"id":"2106.09667","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-06-17T17:20:45Z","cross_cats_sorted":[],"title_canon_sha256":"c986a1824022a058a2f6c64ef8f2ee472a8731600285e502cdd9b09f6de963fa","abstract_canon_sha256":"7421ac9a1c9a80c55768ae7c2fc207940429ab1631f50b13eb749a919b4cc4f6"},"schema_version":"1.0"},"canonical_sha256":"842f06fbf21293bf018c1994e5beac01ff59f96d176ee7834f3da0427492f754","source":{"kind":"arxiv","id":"2106.09667","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2106.09667","created_at":"2026-07-05T04:08:57Z"},{"alias_kind":"arxiv_version","alias_value":"2106.09667v2","created_at":"2026-07-05T04:08:57Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2106.09667","created_at":"2026-07-05T04:08:57Z"},{"alias_kind":"pith_short_12","alias_value":"QQXQN67SCKJ3","created_at":"2026-07-05T04:08:57Z"},{"alias_kind":"pith_short_16","alias_value":"QQXQN67SCKJ36AMM","created_at":"2026-07-05T04:08:57Z"},{"alias_kind":"pith_short_8","alias_value":"QQXQN67S","created_at":"2026-07-05T04:08:57Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2021:QQXQN67SCKJ36AMMDGKOLPVMAH","target":"record","payload":{"canonical_record":{"source":{"id":"2106.09667","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-06-17T17:20:45Z","cross_cats_sorted":[],"title_canon_sha256":"c986a1824022a058a2f6c64ef8f2ee472a8731600285e502cdd9b09f6de963fa","abstract_canon_sha256":"7421ac9a1c9a80c55768ae7c2fc207940429ab1631f50b13eb749a919b4cc4f6"},"schema_version":"1.0"},"canonical_sha256":"842f06fbf21293bf018c1994e5beac01ff59f96d176ee7834f3da0427492f754","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:08:57.481719Z","signature_b64":"4LPNzGWLO/K1u49RDVsZMuj73N3Jb5x12FH7fMuKpTTzg2K3bDDja2zhLjoEG+OUNK0NQ18O12zaJukPbO3xBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"842f06fbf21293bf018c1994e5beac01ff59f96d176ee7834f3da0427492f754","last_reissued_at":"2026-07-05T04:08:57.481273Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:08:57.481273Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2106.09667","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T04:08:57Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"W9CNj5Hr83doLkS6cKqh7gOCELOP9zeD7Sc4mSKkJCPYaA8TKkDKiKNT/5by7ExQihDcnzczsCvRp1kDvD9+Cg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-08T11:06:14.610392Z"},"content_sha256":"1aa982e467be44def2d3679be82734e29cc06f9802071abf27864158af9ab627","schema_version":"1.0","event_id":"sha256:1aa982e467be44def2d3679be82734e29cc06f9802071abf27864158af9ab627"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2021:QQXQN67SCKJ36AMMDGKOLPVMAH","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Poisoning and Backdooring Contrastive Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andreas Terzis, Nicholas Carlini","submitted_at":"2021-06-17T17:20:45Z","abstract_excerpt":"Multimodal contrastive learning methods like CLIP train on noisy and uncurated training datasets. This is cheaper than labeling datasets manually, and even improves out-of-distribution robustness. We show that this practice makes backdoor and poisoning attacks a significant threat. By poisoning just 0.01% of a dataset (e.g., just 300 images of the 3 million-example Conceptual Captions dataset), we can cause the model to misclassify test images by overlaying a small patch. Targeted poisoning attacks, whereby the model misclassifies a particular test input with an adversarially-desired label, ar"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2106.09667","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/2106.09667/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T04:08:57Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"6VUMi5SNdLcc/h3TbFZnhVe0PwtjJ7SyiPOnEzw4J+5tus8EKnQXhntP2bEUk0NCR7TplPRJ0yWbYrYfUw2KCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-08T11:06:14.610901Z"},"content_sha256":"311534fc4d83b455f5fdfa3b692af3c2fec4c4f3b5628599f38e7434aaef57dd","schema_version":"1.0","event_id":"sha256:311534fc4d83b455f5fdfa3b692af3c2fec4c4f3b5628599f38e7434aaef57dd"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/QQXQN67SCKJ36AMMDGKOLPVMAH/bundle.json","state_url":"https://pith.science/pith/QQXQN67SCKJ36AMMDGKOLPVMAH/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/QQXQN67SCKJ36AMMDGKOLPVMAH/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-08T11:06:14Z","links":{"resolver":"https://pith.science/pith/QQXQN67SCKJ36AMMDGKOLPVMAH","bundle":"https://pith.science/pith/QQXQN67SCKJ36AMMDGKOLPVMAH/bundle.json","state":"https://pith.science/pith/QQXQN67SCKJ36AMMDGKOLPVMAH/state.json","well_known_bundle":"https://pith.science/.well-known/pith/QQXQN67SCKJ36AMMDGKOLPVMAH/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2021:QQXQN67SCKJ36AMMDGKOLPVMAH","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"7421ac9a1c9a80c55768ae7c2fc207940429ab1631f50b13eb749a919b4cc4f6","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-06-17T17:20:45Z","title_canon_sha256":"c986a1824022a058a2f6c64ef8f2ee472a8731600285e502cdd9b09f6de963fa"},"schema_version":"1.0","source":{"id":"2106.09667","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2106.09667","created_at":"2026-07-05T04:08:57Z"},{"alias_kind":"arxiv_version","alias_value":"2106.09667v2","created_at":"2026-07-05T04:08:57Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2106.09667","created_at":"2026-07-05T04:08:57Z"},{"alias_kind":"pith_short_12","alias_value":"QQXQN67SCKJ3","created_at":"2026-07-05T04:08:57Z"},{"alias_kind":"pith_short_16","alias_value":"QQXQN67SCKJ36AMM","created_at":"2026-07-05T04:08:57Z"},{"alias_kind":"pith_short_8","alias_value":"QQXQN67S","created_at":"2026-07-05T04:08:57Z"}],"graph_snapshots":[{"event_id":"sha256:311534fc4d83b455f5fdfa3b692af3c2fec4c4f3b5628599f38e7434aaef57dd","target":"graph","created_at":"2026-07-05T04:08:57Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2106.09667/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Multimodal contrastive learning methods like CLIP train on noisy and uncurated training datasets. This is cheaper than labeling datasets manually, and even improves out-of-distribution robustness. We show that this practice makes backdoor and poisoning attacks a significant threat. By poisoning just 0.01% of a dataset (e.g., just 300 images of the 3 million-example Conceptual Captions dataset), we can cause the model to misclassify test images by overlaying a small patch. Targeted poisoning attacks, whereby the model misclassifies a particular test input with an adversarially-desired label, ar","authors_text":"Andreas Terzis, Nicholas Carlini","cross_cats":[],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-06-17T17:20:45Z","title":"Poisoning and Backdooring Contrastive Learning"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2106.09667","kind":"arxiv","version":2},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:1aa982e467be44def2d3679be82734e29cc06f9802071abf27864158af9ab627","target":"record","created_at":"2026-07-05T04:08:57Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"7421ac9a1c9a80c55768ae7c2fc207940429ab1631f50b13eb749a919b4cc4f6","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-06-17T17:20:45Z","title_canon_sha256":"c986a1824022a058a2f6c64ef8f2ee472a8731600285e502cdd9b09f6de963fa"},"schema_version":"1.0","source":{"id":"2106.09667","kind":"arxiv","version":2}},"canonical_sha256":"842f06fbf21293bf018c1994e5beac01ff59f96d176ee7834f3da0427492f754","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"842f06fbf21293bf018c1994e5beac01ff59f96d176ee7834f3da0427492f754","first_computed_at":"2026-07-05T04:08:57.481273Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T04:08:57.481273Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"4LPNzGWLO/K1u49RDVsZMuj73N3Jb5x12FH7fMuKpTTzg2K3bDDja2zhLjoEG+OUNK0NQ18O12zaJukPbO3xBw==","signature_status":"signed_v1","signed_at":"2026-07-05T04:08:57.481719Z","signed_message":"canonical_sha256_bytes"},"source_id":"2106.09667","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:1aa982e467be44def2d3679be82734e29cc06f9802071abf27864158af9ab627","sha256:311534fc4d83b455f5fdfa3b692af3c2fec4c4f3b5628599f38e7434aaef57dd"],"state_sha256":"15072bead48a9cd85b28ea8f047d82c41258866ed0621afa90c39a5680c1a00d"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"SuVHBVZr0qFVzIUXNuhGFQIEhpnkUq+EIxe7/qbq8iXVWBDsvUFbpfMM5Zj/3vd0muDWzSPNhVcCp+TdMiEdDQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-08T11:06:14.616046Z","bundle_sha256":"767acc0c4768cc92a913eeb15c81990b1267976ad2d37dc1ab1fec034d39c947"}}