{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:U36TD6TTK6WNRXYG6WFCH3WKUI","short_pith_number":"pith:U36TD6TT","schema_version":"1.0","canonical_sha256":"a6fd31fa7357acd8df06f58a23eecaa21c4dc0654256bf0e744aad2148e1559e","source":{"kind":"arxiv","id":"1909.11764","version":5},"attestation_state":"computed","paper":{"title":"FreeLB: Enhanced Adversarial Training for Natural Language Understanding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Chen Zhu, Jingjing Liu, Siqi Sun, Tom Goldstein, Yu Cheng, Zhe Gan","submitted_at":"2019-09-25T20:50:32Z","abstract_excerpt":"Adversarial training, which minimizes the maximal risk for label-preserving input perturbations, has proved to be effective for improving the generalization of language models. In this work, we propose a novel adversarial training algorithm, FreeLB, that promotes higher invariance in the embedding space, by adding adversarial perturbations to word embeddings and minimizing the resultant adversarial risk inside different regions around input samples. To validate the effectiveness of the proposed approach, we apply it to Transformer-based models for natural language understanding and commonsense"},"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":"1909.11764","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2019-09-25T20:50:32Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"e674ff3bbd667b88253c1588e23bece38256238e592a642b2b8a5ddd9eecb208","abstract_canon_sha256":"4eb7299c3861316082cec2d6e6636d0dfb7256b9b78901410e9b843dce99fb81"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:57:34.085035Z","signature_b64":"iHBE8MWertBVqFtu1rQexlohIiJP07Q1TSXstregO5I9gjV+52ahotJ6WP+mFz0CRp6oqgZYkov/yEPpaKGQCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a6fd31fa7357acd8df06f58a23eecaa21c4dc0654256bf0e744aad2148e1559e","last_reissued_at":"2026-07-05T00:57:34.084619Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:57:34.084619Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FreeLB: Enhanced Adversarial Training for Natural Language Understanding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Chen Zhu, Jingjing Liu, Siqi Sun, Tom Goldstein, Yu Cheng, Zhe Gan","submitted_at":"2019-09-25T20:50:32Z","abstract_excerpt":"Adversarial training, which minimizes the maximal risk for label-preserving input perturbations, has proved to be effective for improving the generalization of language models. In this work, we propose a novel adversarial training algorithm, FreeLB, that promotes higher invariance in the embedding space, by adding adversarial perturbations to word embeddings and minimizing the resultant adversarial risk inside different regions around input samples. To validate the effectiveness of the proposed approach, we apply it to Transformer-based models for natural language understanding and commonsense"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1909.11764","kind":"arxiv","version":5},"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/1909.11764/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":"1909.11764","created_at":"2026-07-05T00:57:34.084675+00:00"},{"alias_kind":"arxiv_version","alias_value":"1909.11764v5","created_at":"2026-07-05T00:57:34.084675+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1909.11764","created_at":"2026-07-05T00:57:34.084675+00:00"},{"alias_kind":"pith_short_12","alias_value":"U36TD6TTK6WN","created_at":"2026-07-05T00:57:34.084675+00:00"},{"alias_kind":"pith_short_16","alias_value":"U36TD6TTK6WNRXYG","created_at":"2026-07-05T00:57:34.084675+00:00"},{"alias_kind":"pith_short_8","alias_value":"U36TD6TT","created_at":"2026-07-05T00:57:34.084675+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.08893","citing_title":"Cheap Reward Hacking Detection","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23171","citing_title":"Understanding and Improving Noisy Embedding Techniques in Instruction Finetuning","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2309.00614","citing_title":"Baseline Defenses for Adversarial Attacks Against Aligned Language Models","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"1910.10683","citing_title":"Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer","ref_index":83,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17396","citing_title":"Representation-Guided Parameter-Efficient LLM Unlearning","ref_index":62,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/U36TD6TTK6WNRXYG6WFCH3WKUI","json":"https://pith.science/pith/U36TD6TTK6WNRXYG6WFCH3WKUI.json","graph_json":"https://pith.science/api/pith-number/U36TD6TTK6WNRXYG6WFCH3WKUI/graph.json","events_json":"https://pith.science/api/pith-number/U36TD6TTK6WNRXYG6WFCH3WKUI/events.json","paper":"https://pith.science/paper/U36TD6TT"},"agent_actions":{"view_html":"https://pith.science/pith/U36TD6TTK6WNRXYG6WFCH3WKUI","download_json":"https://pith.science/pith/U36TD6TTK6WNRXYG6WFCH3WKUI.json","view_paper":"https://pith.science/paper/U36TD6TT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1909.11764&json=true","fetch_graph":"https://pith.science/api/pith-number/U36TD6TTK6WNRXYG6WFCH3WKUI/graph.json","fetch_events":"https://pith.science/api/pith-number/U36TD6TTK6WNRXYG6WFCH3WKUI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/U36TD6TTK6WNRXYG6WFCH3WKUI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/U36TD6TTK6WNRXYG6WFCH3WKUI/action/storage_attestation","attest_author":"https://pith.science/pith/U36TD6TTK6WNRXYG6WFCH3WKUI/action/author_attestation","sign_citation":"https://pith.science/pith/U36TD6TTK6WNRXYG6WFCH3WKUI/action/citation_signature","submit_replication":"https://pith.science/pith/U36TD6TTK6WNRXYG6WFCH3WKUI/action/replication_record"}},"created_at":"2026-07-05T00:57:34.084675+00:00","updated_at":"2026-07-05T00:57:34.084675+00:00"}