{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:RPL6HPVDWAAAJY4F4N2FO72Y4Q","short_pith_number":"pith:RPL6HPVD","schema_version":"1.0","canonical_sha256":"8bd7e3bea3b00004e385e374577f58e40b8fd4db1c1ae57043e76a7793fb4931","source":{"kind":"arxiv","id":"2112.06837","version":1},"attestation_state":"computed","paper":{"title":"Sparse Interventions in Language Models with Differentiable Masking","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Dieuwke Hupkes, Ivan Titov, Leon Schmid, Nicola De Cao","submitted_at":"2021-12-13T17:49:16Z","abstract_excerpt":"There has been a lot of interest in understanding what information is captured by hidden representations of language models (LMs). Typically, interpretation methods i) do not guarantee that the model actually uses the encoded information, and ii) do not discover small subsets of neurons responsible for a considered phenomenon. Inspired by causal mediation analysis, we propose a method that discovers within a neural LM a small subset of neurons responsible for a particular linguistic phenomenon, i.e., subsets causing a change in the corresponding token emission probabilities. We use a different"},"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":"2112.06837","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-12-13T17:49:16Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"652954a4b59c126841e6cd5eb7d3c8e7e8ad00134088c361f8809dad9fb63344","abstract_canon_sha256":"d51ea405bf11ababeba69732bf359e44c4910e036538b79a5e51332d0cf4c36b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:40:17.543584Z","signature_b64":"gwR1t9UrCgpWbZaL9vNT6J1Va4lMvwTq4mHy62/Zc3qx+EOcFXsB+H9beMIg8rgbq89hhVeyuJXXTJb/DwC3Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8bd7e3bea3b00004e385e374577f58e40b8fd4db1c1ae57043e76a7793fb4931","last_reissued_at":"2026-07-05T03:40:17.543207Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:40:17.543207Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Sparse Interventions in Language Models with Differentiable Masking","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Dieuwke Hupkes, Ivan Titov, Leon Schmid, Nicola De Cao","submitted_at":"2021-12-13T17:49:16Z","abstract_excerpt":"There has been a lot of interest in understanding what information is captured by hidden representations of language models (LMs). Typically, interpretation methods i) do not guarantee that the model actually uses the encoded information, and ii) do not discover small subsets of neurons responsible for a considered phenomenon. Inspired by causal mediation analysis, we propose a method that discovers within a neural LM a small subset of neurons responsible for a particular linguistic phenomenon, i.e., subsets causing a change in the corresponding token emission probabilities. We use a different"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2112.06837","kind":"arxiv","version":1},"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/2112.06837/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":"2112.06837","created_at":"2026-07-05T03:40:17.543257+00:00"},{"alias_kind":"arxiv_version","alias_value":"2112.06837v1","created_at":"2026-07-05T03:40:17.543257+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2112.06837","created_at":"2026-07-05T03:40:17.543257+00:00"},{"alias_kind":"pith_short_12","alias_value":"RPL6HPVDWAAA","created_at":"2026-07-05T03:40:17.543257+00:00"},{"alias_kind":"pith_short_16","alias_value":"RPL6HPVDWAAAJY4F","created_at":"2026-07-05T03:40:17.543257+00:00"},{"alias_kind":"pith_short_8","alias_value":"RPL6HPVD","created_at":"2026-07-05T03:40:17.543257+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2501.14926","citing_title":"Interpretability in Parameter Space: Minimizing Mechanistic Description Length with Attribution-based Parameter Decomposition","ref_index":9,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RPL6HPVDWAAAJY4F4N2FO72Y4Q","json":"https://pith.science/pith/RPL6HPVDWAAAJY4F4N2FO72Y4Q.json","graph_json":"https://pith.science/api/pith-number/RPL6HPVDWAAAJY4F4N2FO72Y4Q/graph.json","events_json":"https://pith.science/api/pith-number/RPL6HPVDWAAAJY4F4N2FO72Y4Q/events.json","paper":"https://pith.science/paper/RPL6HPVD"},"agent_actions":{"view_html":"https://pith.science/pith/RPL6HPVDWAAAJY4F4N2FO72Y4Q","download_json":"https://pith.science/pith/RPL6HPVDWAAAJY4F4N2FO72Y4Q.json","view_paper":"https://pith.science/paper/RPL6HPVD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2112.06837&json=true","fetch_graph":"https://pith.science/api/pith-number/RPL6HPVDWAAAJY4F4N2FO72Y4Q/graph.json","fetch_events":"https://pith.science/api/pith-number/RPL6HPVDWAAAJY4F4N2FO72Y4Q/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RPL6HPVDWAAAJY4F4N2FO72Y4Q/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RPL6HPVDWAAAJY4F4N2FO72Y4Q/action/storage_attestation","attest_author":"https://pith.science/pith/RPL6HPVDWAAAJY4F4N2FO72Y4Q/action/author_attestation","sign_citation":"https://pith.science/pith/RPL6HPVDWAAAJY4F4N2FO72Y4Q/action/citation_signature","submit_replication":"https://pith.science/pith/RPL6HPVDWAAAJY4F4N2FO72Y4Q/action/replication_record"}},"created_at":"2026-07-05T03:40:17.543257+00:00","updated_at":"2026-07-05T03:40:17.543257+00:00"}