{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:XGLJN6HVSZ2OVTMNOUS27POGDY","short_pith_number":"pith:XGLJN6HV","schema_version":"1.0","canonical_sha256":"b99696f8f59674eacd8d7525afbdc61e0e9d603328a9ed6b4f1ffb14a483dea7","source":{"kind":"arxiv","id":"2112.00826","version":2},"attestation_state":"computed","paper":{"title":"Inducing Causal Structure for Interpretable Neural Networks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Atticus Geiger, Christopher Potts, Elisa Kreiss, Hanson Lu, Josh Rozner, Noah D. Goodman, Thomas Icard, Zhengxuan Wu","submitted_at":"2021-12-01T21:07:01Z","abstract_excerpt":"In many areas, we have well-founded insights about causal structure that would be useful to bring into our trained models while still allowing them to learn in a data-driven fashion. To achieve this, we present the new method of interchange intervention training (IIT). In IIT, we (1) align variables in a causal model (e.g., a deterministic program or Bayesian network) with representations in a neural model and (2) train the neural model to match the counterfactual behavior of the causal model on a base input when aligned representations in both models are set to be the value they would be for "},"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":"2112.00826","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2021-12-01T21:07:01Z","cross_cats_sorted":[],"title_canon_sha256":"087bed6f85eca90122a7c11b37f386dd973a27d94c5d3884924fc1f02e59e492","abstract_canon_sha256":"c99fc8cd4520a3117790264e49ca6127f418052417613fde949c88e8619ef632"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:42:15.483581Z","signature_b64":"71KxpbI502/FO+SL3a6wc3WhrEqcJYHq0ERiG7p5rdB8FECjVgp7h6P2rjFrj55BhxjceZ8iXc2B8tF/8+lDDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b99696f8f59674eacd8d7525afbdc61e0e9d603328a9ed6b4f1ffb14a483dea7","last_reissued_at":"2026-07-05T04:42:15.483142Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:42:15.483142Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Inducing Causal Structure for Interpretable Neural Networks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Atticus Geiger, Christopher Potts, Elisa Kreiss, Hanson Lu, Josh Rozner, Noah D. Goodman, Thomas Icard, Zhengxuan Wu","submitted_at":"2021-12-01T21:07:01Z","abstract_excerpt":"In many areas, we have well-founded insights about causal structure that would be useful to bring into our trained models while still allowing them to learn in a data-driven fashion. To achieve this, we present the new method of interchange intervention training (IIT). In IIT, we (1) align variables in a causal model (e.g., a deterministic program or Bayesian network) with representations in a neural model and (2) train the neural model to match the counterfactual behavior of the causal model on a base input when aligned representations in both models are set to be the value they would be for "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2112.00826","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/2112.00826/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":"2112.00826","created_at":"2026-07-05T04:42:15.483198+00:00"},{"alias_kind":"arxiv_version","alias_value":"2112.00826v2","created_at":"2026-07-05T04:42:15.483198+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2112.00826","created_at":"2026-07-05T04:42:15.483198+00:00"},{"alias_kind":"pith_short_12","alias_value":"XGLJN6HVSZ2O","created_at":"2026-07-05T04:42:15.483198+00:00"},{"alias_kind":"pith_short_16","alias_value":"XGLJN6HVSZ2OVTMN","created_at":"2026-07-05T04:42:15.483198+00:00"},{"alias_kind":"pith_short_8","alias_value":"XGLJN6HV","created_at":"2026-07-05T04:42:15.483198+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":260,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XGLJN6HVSZ2OVTMNOUS27POGDY","json":"https://pith.science/pith/XGLJN6HVSZ2OVTMNOUS27POGDY.json","graph_json":"https://pith.science/api/pith-number/XGLJN6HVSZ2OVTMNOUS27POGDY/graph.json","events_json":"https://pith.science/api/pith-number/XGLJN6HVSZ2OVTMNOUS27POGDY/events.json","paper":"https://pith.science/paper/XGLJN6HV"},"agent_actions":{"view_html":"https://pith.science/pith/XGLJN6HVSZ2OVTMNOUS27POGDY","download_json":"https://pith.science/pith/XGLJN6HVSZ2OVTMNOUS27POGDY.json","view_paper":"https://pith.science/paper/XGLJN6HV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2112.00826&json=true","fetch_graph":"https://pith.science/api/pith-number/XGLJN6HVSZ2OVTMNOUS27POGDY/graph.json","fetch_events":"https://pith.science/api/pith-number/XGLJN6HVSZ2OVTMNOUS27POGDY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XGLJN6HVSZ2OVTMNOUS27POGDY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XGLJN6HVSZ2OVTMNOUS27POGDY/action/storage_attestation","attest_author":"https://pith.science/pith/XGLJN6HVSZ2OVTMNOUS27POGDY/action/author_attestation","sign_citation":"https://pith.science/pith/XGLJN6HVSZ2OVTMNOUS27POGDY/action/citation_signature","submit_replication":"https://pith.science/pith/XGLJN6HVSZ2OVTMNOUS27POGDY/action/replication_record"}},"created_at":"2026-07-05T04:42:15.483198+00:00","updated_at":"2026-07-05T04:42:15.483198+00:00"}