{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:AFJ7JZKSECR6XOZQE4RMYOVH7W","short_pith_number":"pith:AFJ7JZKS","schema_version":"1.0","canonical_sha256":"0153f4e55220a3ebbb302722cc3aa7fd9fa260ae010656ead1266ad272a81180","source":{"kind":"arxiv","id":"2503.01824","version":1},"attestation_state":"computed","paper":{"title":"From superposition to sparse codes: interpretable representations in neural networks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Charles O'Neill, David Klindt, Harald Maurer, Nina Miolane, Patrik Reizinger","submitted_at":"2025-03-03T18:49:59Z","abstract_excerpt":"Understanding how information is represented in neural networks is a fundamental challenge in both neuroscience and artificial intelligence. Despite their nonlinear architectures, recent evidence suggests that neural networks encode features in superposition, meaning that input concepts are linearly overlaid within the network's representations. We present a perspective that explains this phenomenon and provides a foundation for extracting interpretable representations from neural activations. Our theoretical framework consists of three steps: (1) Identifiability theory shows that neural netwo"},"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":"2503.01824","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-03-03T18:49:59Z","cross_cats_sorted":[],"title_canon_sha256":"1873c68ce9342c3d2cab611880f1c80b0045b30d89310352d6af637a2f380dbb","abstract_canon_sha256":"5dbe3f100bcdd41b746a48d1edfde21c37bc85c1f3f8a3c648ed6f8cd8bddf73"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:23:16.183234Z","signature_b64":"aGUBy+1vKdL91m7YiGZv/N+2meJY9TYcXXoWwwLHMtMKfPTxhrr0hZgm+n9MMth1fTQ2FpIkJhzgH1Gnoa1bBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0153f4e55220a3ebbb302722cc3aa7fd9fa260ae010656ead1266ad272a81180","last_reissued_at":"2026-07-05T10:23:16.182509Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:23:16.182509Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"From superposition to sparse codes: interpretable representations in neural networks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Charles O'Neill, David Klindt, Harald Maurer, Nina Miolane, Patrik Reizinger","submitted_at":"2025-03-03T18:49:59Z","abstract_excerpt":"Understanding how information is represented in neural networks is a fundamental challenge in both neuroscience and artificial intelligence. Despite their nonlinear architectures, recent evidence suggests that neural networks encode features in superposition, meaning that input concepts are linearly overlaid within the network's representations. We present a perspective that explains this phenomenon and provides a foundation for extracting interpretable representations from neural activations. Our theoretical framework consists of three steps: (1) Identifiability theory shows that neural netwo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.01824","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/2503.01824/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":"2503.01824","created_at":"2026-07-05T10:23:16.182596+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.01824v1","created_at":"2026-07-05T10:23:16.182596+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.01824","created_at":"2026-07-05T10:23:16.182596+00:00"},{"alias_kind":"pith_short_12","alias_value":"AFJ7JZKSECR6","created_at":"2026-07-05T10:23:16.182596+00:00"},{"alias_kind":"pith_short_16","alias_value":"AFJ7JZKSECR6XOZQ","created_at":"2026-07-05T10:23:16.182596+00:00"},{"alias_kind":"pith_short_8","alias_value":"AFJ7JZKS","created_at":"2026-07-05T10:23:16.182596+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25234","citing_title":"Structuring Sparsity: Block-Sparse Featurizers Capture Visual Concept Manifolds","ref_index":71,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01799","citing_title":"Expander Sparse Autoencoders: Parameter-Efficient Dictionaries for Mechanistic Interpretability","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26379","citing_title":"When Does LeJEPA Learn a World Model?","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26921","citing_title":"Similarity-based representation factorization for revealing interpretable dimensions in representational data","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18537","citing_title":"Probing for Representation Manifolds in Superposition","ref_index":86,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AFJ7JZKSECR6XOZQE4RMYOVH7W","json":"https://pith.science/pith/AFJ7JZKSECR6XOZQE4RMYOVH7W.json","graph_json":"https://pith.science/api/pith-number/AFJ7JZKSECR6XOZQE4RMYOVH7W/graph.json","events_json":"https://pith.science/api/pith-number/AFJ7JZKSECR6XOZQE4RMYOVH7W/events.json","paper":"https://pith.science/paper/AFJ7JZKS"},"agent_actions":{"view_html":"https://pith.science/pith/AFJ7JZKSECR6XOZQE4RMYOVH7W","download_json":"https://pith.science/pith/AFJ7JZKSECR6XOZQE4RMYOVH7W.json","view_paper":"https://pith.science/paper/AFJ7JZKS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.01824&json=true","fetch_graph":"https://pith.science/api/pith-number/AFJ7JZKSECR6XOZQE4RMYOVH7W/graph.json","fetch_events":"https://pith.science/api/pith-number/AFJ7JZKSECR6XOZQE4RMYOVH7W/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AFJ7JZKSECR6XOZQE4RMYOVH7W/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AFJ7JZKSECR6XOZQE4RMYOVH7W/action/storage_attestation","attest_author":"https://pith.science/pith/AFJ7JZKSECR6XOZQE4RMYOVH7W/action/author_attestation","sign_citation":"https://pith.science/pith/AFJ7JZKSECR6XOZQE4RMYOVH7W/action/citation_signature","submit_replication":"https://pith.science/pith/AFJ7JZKSECR6XOZQE4RMYOVH7W/action/replication_record"}},"created_at":"2026-07-05T10:23:16.182596+00:00","updated_at":"2026-07-05T10:23:16.182596+00:00"}