{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:6AACUU5C55JQCCPPCYD7O3WTZT","short_pith_number":"pith:6AACUU5C","schema_version":"1.0","canonical_sha256":"f0002a53a2ef530109ef1607f76ed3cceecbbe4f357c3731858c9884d5902246","source":{"kind":"arxiv","id":"2004.11207","version":2},"attestation_state":"computed","paper":{"title":"Self-Attention Attribution: Interpreting Information Interactions Inside Transformer","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Furu Wei, Ke Xu, Li Dong, Yaru Hao","submitted_at":"2020-04-23T14:58:22Z","abstract_excerpt":"The great success of Transformer-based models benefits from the powerful multi-head self-attention mechanism, which learns token dependencies and encodes contextual information from the input. Prior work strives to attribute model decisions to individual input features with different saliency measures, but they fail to explain how these input features interact with each other to reach predictions. In this paper, we propose a self-attention attribution method to interpret the information interactions inside Transformer. We take BERT as an example to conduct extensive studies. Firstly, we apply "},"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":"2004.11207","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-04-23T14:58:22Z","cross_cats_sorted":[],"title_canon_sha256":"559a2312148d0659d437adf4d9e92ecadb46e323c3ab5708e84107b7ed38801f","abstract_canon_sha256":"5a953e81f9cdf1cc1f75dd68695a2feb50ad986f4b471534f143ec88450431c9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:18:15.690220Z","signature_b64":"RPJ7Lt85HnYuP8713dTh/ovn+RSQfl8gJbx9GJJgCZIWFbB+8lL3SXYkRcenrrlAuXYduDEKuBzvvIlJWXw2CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f0002a53a2ef530109ef1607f76ed3cceecbbe4f357c3731858c9884d5902246","last_reissued_at":"2026-07-05T02:18:15.689743Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:18:15.689743Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Self-Attention Attribution: Interpreting Information Interactions Inside Transformer","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Furu Wei, Ke Xu, Li Dong, Yaru Hao","submitted_at":"2020-04-23T14:58:22Z","abstract_excerpt":"The great success of Transformer-based models benefits from the powerful multi-head self-attention mechanism, which learns token dependencies and encodes contextual information from the input. Prior work strives to attribute model decisions to individual input features with different saliency measures, but they fail to explain how these input features interact with each other to reach predictions. In this paper, we propose a self-attention attribution method to interpret the information interactions inside Transformer. We take BERT as an example to conduct extensive studies. Firstly, we apply "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2004.11207","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/2004.11207/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":"2004.11207","created_at":"2026-07-05T02:18:15.689807+00:00"},{"alias_kind":"arxiv_version","alias_value":"2004.11207v2","created_at":"2026-07-05T02:18:15.689807+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2004.11207","created_at":"2026-07-05T02:18:15.689807+00:00"},{"alias_kind":"pith_short_12","alias_value":"6AACUU5C55JQ","created_at":"2026-07-05T02:18:15.689807+00:00"},{"alias_kind":"pith_short_16","alias_value":"6AACUU5C55JQCCPP","created_at":"2026-07-05T02:18:15.689807+00:00"},{"alias_kind":"pith_short_8","alias_value":"6AACUU5C","created_at":"2026-07-05T02:18:15.689807+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2106.08254","citing_title":"BEiT: BERT Pre-Training of Image Transformers","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6AACUU5C55JQCCPPCYD7O3WTZT","json":"https://pith.science/pith/6AACUU5C55JQCCPPCYD7O3WTZT.json","graph_json":"https://pith.science/api/pith-number/6AACUU5C55JQCCPPCYD7O3WTZT/graph.json","events_json":"https://pith.science/api/pith-number/6AACUU5C55JQCCPPCYD7O3WTZT/events.json","paper":"https://pith.science/paper/6AACUU5C"},"agent_actions":{"view_html":"https://pith.science/pith/6AACUU5C55JQCCPPCYD7O3WTZT","download_json":"https://pith.science/pith/6AACUU5C55JQCCPPCYD7O3WTZT.json","view_paper":"https://pith.science/paper/6AACUU5C","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2004.11207&json=true","fetch_graph":"https://pith.science/api/pith-number/6AACUU5C55JQCCPPCYD7O3WTZT/graph.json","fetch_events":"https://pith.science/api/pith-number/6AACUU5C55JQCCPPCYD7O3WTZT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6AACUU5C55JQCCPPCYD7O3WTZT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6AACUU5C55JQCCPPCYD7O3WTZT/action/storage_attestation","attest_author":"https://pith.science/pith/6AACUU5C55JQCCPPCYD7O3WTZT/action/author_attestation","sign_citation":"https://pith.science/pith/6AACUU5C55JQCCPPCYD7O3WTZT/action/citation_signature","submit_replication":"https://pith.science/pith/6AACUU5C55JQCCPPCYD7O3WTZT/action/replication_record"}},"created_at":"2026-07-05T02:18:15.689807+00:00","updated_at":"2026-07-05T02:18:15.689807+00:00"}