{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:4JBRKML4XQLVXWJYWUFMJBWH34","short_pith_number":"pith:4JBRKML4","schema_version":"1.0","canonical_sha256":"e24315317cbc175bd938b50ac486c7df3ad3bf95d4a07e3d22d8a61eba199f07","source":{"kind":"arxiv","id":"2410.11767","version":4},"attestation_state":"computed","paper":{"title":"Analyzing (In)Abilities of SAEs via Formal Languages","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Abhinav Menon, David Krueger, Ekdeep Singh Lubana, Manish Shrivastava","submitted_at":"2024-10-15T16:42:13Z","abstract_excerpt":"Autoencoders have been used for finding interpretable and disentangled features underlying neural network representations in both image and text domains. While the efficacy and pitfalls of such methods are well-studied in vision, there is a lack of corresponding results, both qualitative and quantitative, for the text domain. We aim to address this gap by training sparse autoencoders (SAEs) on a synthetic testbed of formal languages. Specifically, we train SAEs on the hidden representations of models trained on formal languages (Dyck-2, Expr, and English PCFG) under a wide variety of hyperpara"},"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":"2410.11767","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-10-15T16:42:13Z","cross_cats_sorted":[],"title_canon_sha256":"209971c57fde2a74e93b7b1966f022d9605b7009c075d2823c55e8baf309803f","abstract_canon_sha256":"14e0bdcb8c6d058bc6871bc8c33a3f4876aba52629b71029041fe3405e9775a6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:09:48.997775Z","signature_b64":"4fT4qZXQVoy4TsC8Q633CsZgNWV82na0MZPHLlZBDHX6QEuE/9KnfRZ8eeTlgC5ibPSYSeFpF8vMPNTYBpdHDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e24315317cbc175bd938b50ac486c7df3ad3bf95d4a07e3d22d8a61eba199f07","last_reissued_at":"2026-07-05T10:09:48.997412Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:09:48.997412Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Analyzing (In)Abilities of SAEs via Formal Languages","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Abhinav Menon, David Krueger, Ekdeep Singh Lubana, Manish Shrivastava","submitted_at":"2024-10-15T16:42:13Z","abstract_excerpt":"Autoencoders have been used for finding interpretable and disentangled features underlying neural network representations in both image and text domains. While the efficacy and pitfalls of such methods are well-studied in vision, there is a lack of corresponding results, both qualitative and quantitative, for the text domain. We aim to address this gap by training sparse autoencoders (SAEs) on a synthetic testbed of formal languages. Specifically, we train SAEs on the hidden representations of models trained on formal languages (Dyck-2, Expr, and English PCFG) under a wide variety of hyperpara"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.11767","kind":"arxiv","version":4},"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/2410.11767/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":"2410.11767","created_at":"2026-07-05T10:09:48.997468+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.11767v4","created_at":"2026-07-05T10:09:48.997468+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.11767","created_at":"2026-07-05T10:09:48.997468+00:00"},{"alias_kind":"pith_short_12","alias_value":"4JBRKML4XQLV","created_at":"2026-07-05T10:09:48.997468+00:00"},{"alias_kind":"pith_short_16","alias_value":"4JBRKML4XQLVXWJY","created_at":"2026-07-05T10:09:48.997468+00:00"},{"alias_kind":"pith_short_8","alias_value":"4JBRKML4","created_at":"2026-07-05T10:09:48.997468+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4JBRKML4XQLVXWJYWUFMJBWH34","json":"https://pith.science/pith/4JBRKML4XQLVXWJYWUFMJBWH34.json","graph_json":"https://pith.science/api/pith-number/4JBRKML4XQLVXWJYWUFMJBWH34/graph.json","events_json":"https://pith.science/api/pith-number/4JBRKML4XQLVXWJYWUFMJBWH34/events.json","paper":"https://pith.science/paper/4JBRKML4"},"agent_actions":{"view_html":"https://pith.science/pith/4JBRKML4XQLVXWJYWUFMJBWH34","download_json":"https://pith.science/pith/4JBRKML4XQLVXWJYWUFMJBWH34.json","view_paper":"https://pith.science/paper/4JBRKML4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.11767&json=true","fetch_graph":"https://pith.science/api/pith-number/4JBRKML4XQLVXWJYWUFMJBWH34/graph.json","fetch_events":"https://pith.science/api/pith-number/4JBRKML4XQLVXWJYWUFMJBWH34/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4JBRKML4XQLVXWJYWUFMJBWH34/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4JBRKML4XQLVXWJYWUFMJBWH34/action/storage_attestation","attest_author":"https://pith.science/pith/4JBRKML4XQLVXWJYWUFMJBWH34/action/author_attestation","sign_citation":"https://pith.science/pith/4JBRKML4XQLVXWJYWUFMJBWH34/action/citation_signature","submit_replication":"https://pith.science/pith/4JBRKML4XQLVXWJYWUFMJBWH34/action/replication_record"}},"created_at":"2026-07-05T10:09:48.997468+00:00","updated_at":"2026-07-05T10:09:48.997468+00:00"}