{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:V3CQYRPMCQBH3OZZI6L73QHJMC","short_pith_number":"pith:V3CQYRPM","schema_version":"1.0","canonical_sha256":"aec50c45ec14027dbb394797fdc0e960b58ee0c16d8c7715ea9827adbe2cc992","source":{"kind":"arxiv","id":"2504.00194","version":1},"attestation_state":"computed","paper":{"title":"Identifying Sparsely Active Circuits Through Local Loss Landscape Decomposition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Brianna Chrisman, Lee Sharkey, Lucius Bushnaq","submitted_at":"2025-03-31T20:04:39Z","abstract_excerpt":"Much of mechanistic interpretability has focused on understanding the activation spaces of large neural networks. However, activation space-based approaches reveal little about the underlying circuitry used to compute features. To better understand the circuits employed by models, we introduce a new decomposition method called Local Loss Landscape Decomposition (L3D). L3D identifies a set of low-rank subnetworks: directions in parameter space of which a subset can reconstruct the gradient of the loss between any sample's output and a reference output vector. We design a series of progressively"},"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":"2504.00194","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-03-31T20:04:39Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"df55b93a246cb1a54e39b0cb478f662856e2435501f5c390e70298588637b0f7","abstract_canon_sha256":"f77f985f30255bb895302dbf9f6445e8383339916f5576b944571c34920bb004"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:42:21.710096Z","signature_b64":"WJWFIu3dUc1NweJVNBsuZ/6dnXccAHNpeYKKL6GzAjENE4jiFr9DZEyvUNF6iHo7biIGFlbkRqFddEE9NN8lCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"aec50c45ec14027dbb394797fdc0e960b58ee0c16d8c7715ea9827adbe2cc992","last_reissued_at":"2026-07-05T10:42:21.709641Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:42:21.709641Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Identifying Sparsely Active Circuits Through Local Loss Landscape Decomposition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Brianna Chrisman, Lee Sharkey, Lucius Bushnaq","submitted_at":"2025-03-31T20:04:39Z","abstract_excerpt":"Much of mechanistic interpretability has focused on understanding the activation spaces of large neural networks. However, activation space-based approaches reveal little about the underlying circuitry used to compute features. To better understand the circuits employed by models, we introduce a new decomposition method called Local Loss Landscape Decomposition (L3D). L3D identifies a set of low-rank subnetworks: directions in parameter space of which a subset can reconstruct the gradient of the loss between any sample's output and a reference output vector. We design a series of progressively"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.00194","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/2504.00194/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":"2504.00194","created_at":"2026-07-05T10:42:21.709699+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.00194v1","created_at":"2026-07-05T10:42:21.709699+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.00194","created_at":"2026-07-05T10:42:21.709699+00:00"},{"alias_kind":"pith_short_12","alias_value":"V3CQYRPMCQBH","created_at":"2026-07-05T10:42:21.709699+00:00"},{"alias_kind":"pith_short_16","alias_value":"V3CQYRPMCQBH3OZZ","created_at":"2026-07-05T10:42:21.709699+00:00"},{"alias_kind":"pith_short_8","alias_value":"V3CQYRPM","created_at":"2026-07-05T10:42:21.709699+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.20790","citing_title":"Stochastic Parameter Decomposition","ref_index":8,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/V3CQYRPMCQBH3OZZI6L73QHJMC","json":"https://pith.science/pith/V3CQYRPMCQBH3OZZI6L73QHJMC.json","graph_json":"https://pith.science/api/pith-number/V3CQYRPMCQBH3OZZI6L73QHJMC/graph.json","events_json":"https://pith.science/api/pith-number/V3CQYRPMCQBH3OZZI6L73QHJMC/events.json","paper":"https://pith.science/paper/V3CQYRPM"},"agent_actions":{"view_html":"https://pith.science/pith/V3CQYRPMCQBH3OZZI6L73QHJMC","download_json":"https://pith.science/pith/V3CQYRPMCQBH3OZZI6L73QHJMC.json","view_paper":"https://pith.science/paper/V3CQYRPM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.00194&json=true","fetch_graph":"https://pith.science/api/pith-number/V3CQYRPMCQBH3OZZI6L73QHJMC/graph.json","fetch_events":"https://pith.science/api/pith-number/V3CQYRPMCQBH3OZZI6L73QHJMC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/V3CQYRPMCQBH3OZZI6L73QHJMC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/V3CQYRPMCQBH3OZZI6L73QHJMC/action/storage_attestation","attest_author":"https://pith.science/pith/V3CQYRPMCQBH3OZZI6L73QHJMC/action/author_attestation","sign_citation":"https://pith.science/pith/V3CQYRPMCQBH3OZZI6L73QHJMC/action/citation_signature","submit_replication":"https://pith.science/pith/V3CQYRPMCQBH3OZZI6L73QHJMC/action/replication_record"}},"created_at":"2026-07-05T10:42:21.709699+00:00","updated_at":"2026-07-05T10:42:21.709699+00:00"}