{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:LAEQKSHBQHC7UPQYVDLE3CTXVR","short_pith_number":"pith:LAEQKSHB","schema_version":"1.0","canonical_sha256":"58090548e181c5fa3e18a8d64d8a77ac460548367a7899d48fc9f1f3a68167f9","source":{"kind":"arxiv","id":"2104.08678","version":3},"attestation_state":"computed","paper":{"title":"Improving Question Answering Model Robustness with Synthetic Adversarial Data Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Douwe Kiela, Max Bartolo, Pontus Stenetorp, Robin Jia, Sebastian Riedel, Tristan Thrush","submitted_at":"2021-04-18T02:00:06Z","abstract_excerpt":"Despite recent progress, state-of-the-art question answering models remain vulnerable to a variety of adversarial attacks. While dynamic adversarial data collection, in which a human annotator tries to write examples that fool a model-in-the-loop, can improve model robustness, this process is expensive which limits the scale of the collected data. In this work, we are the first to use synthetic adversarial data generation to make question answering models more robust to human adversaries. We develop a data generation pipeline that selects source passages, identifies candidate answers, generate"},"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":"2104.08678","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2021-04-18T02:00:06Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"30f3ddf8a7c18d5c33de26d502459a244511f50b0a9c0d02dd9f238a51b6eca7","abstract_canon_sha256":"be3e1baaef96e1620a5ecabe695828b4149c13d411f5a3134b1ab9f0b3dbedb4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:05:11.275460Z","signature_b64":"LdFsmzRRt9PI+5MfZn6kBqSpdRtRNl8N2cTkA03yOEH69Hn/3ywLYaG2q4M1WS2oTfAIWvKVB2N0iicOptWGDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"58090548e181c5fa3e18a8d64d8a77ac460548367a7899d48fc9f1f3a68167f9","last_reissued_at":"2026-07-05T04:05:11.274932Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:05:11.274932Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Improving Question Answering Model Robustness with Synthetic Adversarial Data Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Douwe Kiela, Max Bartolo, Pontus Stenetorp, Robin Jia, Sebastian Riedel, Tristan Thrush","submitted_at":"2021-04-18T02:00:06Z","abstract_excerpt":"Despite recent progress, state-of-the-art question answering models remain vulnerable to a variety of adversarial attacks. While dynamic adversarial data collection, in which a human annotator tries to write examples that fool a model-in-the-loop, can improve model robustness, this process is expensive which limits the scale of the collected data. In this work, we are the first to use synthetic adversarial data generation to make question answering models more robust to human adversaries. We develop a data generation pipeline that selects source passages, identifies candidate answers, generate"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2104.08678","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/2104.08678/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":"2104.08678","created_at":"2026-07-05T04:05:11.274993+00:00"},{"alias_kind":"arxiv_version","alias_value":"2104.08678v3","created_at":"2026-07-05T04:05:11.274993+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2104.08678","created_at":"2026-07-05T04:05:11.274993+00:00"},{"alias_kind":"pith_short_12","alias_value":"LAEQKSHBQHC7","created_at":"2026-07-05T04:05:11.274993+00:00"},{"alias_kind":"pith_short_16","alias_value":"LAEQKSHBQHC7UPQY","created_at":"2026-07-05T04:05:11.274993+00:00"},{"alias_kind":"pith_short_8","alias_value":"LAEQKSHB","created_at":"2026-07-05T04:05:11.274993+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2406.09250","citing_title":"MirrorCheck: Efficient Adversarial Defense for Vision-Language Models","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2406.10162","citing_title":"Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models","ref_index":96,"is_internal_anchor":false},{"citing_arxiv_id":"2303.09014","citing_title":"ART: Automatic multi-step reasoning and tool-use for large language models","ref_index":81,"is_internal_anchor":false},{"citing_arxiv_id":"2310.08419","citing_title":"Jailbreaking Black Box Large Language Models in Twenty Queries","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2209.07858","citing_title":"Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LAEQKSHBQHC7UPQYVDLE3CTXVR","json":"https://pith.science/pith/LAEQKSHBQHC7UPQYVDLE3CTXVR.json","graph_json":"https://pith.science/api/pith-number/LAEQKSHBQHC7UPQYVDLE3CTXVR/graph.json","events_json":"https://pith.science/api/pith-number/LAEQKSHBQHC7UPQYVDLE3CTXVR/events.json","paper":"https://pith.science/paper/LAEQKSHB"},"agent_actions":{"view_html":"https://pith.science/pith/LAEQKSHBQHC7UPQYVDLE3CTXVR","download_json":"https://pith.science/pith/LAEQKSHBQHC7UPQYVDLE3CTXVR.json","view_paper":"https://pith.science/paper/LAEQKSHB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2104.08678&json=true","fetch_graph":"https://pith.science/api/pith-number/LAEQKSHBQHC7UPQYVDLE3CTXVR/graph.json","fetch_events":"https://pith.science/api/pith-number/LAEQKSHBQHC7UPQYVDLE3CTXVR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LAEQKSHBQHC7UPQYVDLE3CTXVR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LAEQKSHBQHC7UPQYVDLE3CTXVR/action/storage_attestation","attest_author":"https://pith.science/pith/LAEQKSHBQHC7UPQYVDLE3CTXVR/action/author_attestation","sign_citation":"https://pith.science/pith/LAEQKSHBQHC7UPQYVDLE3CTXVR/action/citation_signature","submit_replication":"https://pith.science/pith/LAEQKSHBQHC7UPQYVDLE3CTXVR/action/replication_record"}},"created_at":"2026-07-05T04:05:11.274993+00:00","updated_at":"2026-07-05T04:05:11.274993+00:00"}