{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:NHE6LKZ2NNH66LELCBN3ZFUM6B","short_pith_number":"pith:NHE6LKZ2","schema_version":"1.0","canonical_sha256":"69c9e5ab3a6b4fef2c8b105bbc968cf0515ca20dffa9dbbc164667115753ac0e","source":{"kind":"arxiv","id":"2103.15949","version":2},"attestation_state":"computed","paper":{"title":"Transformer visualization via dictionary learning: contextualized embedding as a linear superposition of transformer factors","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bruno A Olshausen, Yann LeCun, Yubei Chen, Zeyu Yun","submitted_at":"2021-03-29T20:51:33Z","abstract_excerpt":"Transformer networks have revolutionized NLP representation learning since they were introduced. Though a great effort has been made to explain the representation in transformers, it is widely recognized that our understanding is not sufficient. One important reason is that there lack enough visualization tools for detailed analysis. In this paper, we propose to use dictionary learning to open up these \"black boxes\" as linear superpositions of transformer factors. Through visualization, we demonstrate the hierarchical semantic structures captured by the transformer factors, e.g., word-level po"},"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":"2103.15949","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-03-29T20:51:33Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"ff976b83718eb09af57a0c8595de7ad455ed1d56b359e2bdd36b08440eec9a3a","abstract_canon_sha256":"6abf56e670c99aae761c02f7b6564bd4fe3bef3dbfe9b25b0a5c9ec6a12642d1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:57:28.798918Z","signature_b64":"/ps88whfSuy0ODOqOygdpb5Rd8E3PxaI3BJhUISUCXuQ/FmSI0mr2UnPGw3p13RdahFwLOCr5t79oxeTSL9mAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"69c9e5ab3a6b4fef2c8b105bbc968cf0515ca20dffa9dbbc164667115753ac0e","last_reissued_at":"2026-07-05T05:57:28.798493Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:57:28.798493Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Transformer visualization via dictionary learning: contextualized embedding as a linear superposition of transformer factors","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bruno A Olshausen, Yann LeCun, Yubei Chen, Zeyu Yun","submitted_at":"2021-03-29T20:51:33Z","abstract_excerpt":"Transformer networks have revolutionized NLP representation learning since they were introduced. Though a great effort has been made to explain the representation in transformers, it is widely recognized that our understanding is not sufficient. One important reason is that there lack enough visualization tools for detailed analysis. In this paper, we propose to use dictionary learning to open up these \"black boxes\" as linear superpositions of transformer factors. Through visualization, we demonstrate the hierarchical semantic structures captured by the transformer factors, e.g., word-level po"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2103.15949","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/2103.15949/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":"2103.15949","created_at":"2026-07-05T05:57:28.798549+00:00"},{"alias_kind":"arxiv_version","alias_value":"2103.15949v2","created_at":"2026-07-05T05:57:28.798549+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2103.15949","created_at":"2026-07-05T05:57:28.798549+00:00"},{"alias_kind":"pith_short_12","alias_value":"NHE6LKZ2NNH6","created_at":"2026-07-05T05:57:28.798549+00:00"},{"alias_kind":"pith_short_16","alias_value":"NHE6LKZ2NNH66LEL","created_at":"2026-07-05T05:57:28.798549+00:00"},{"alias_kind":"pith_short_8","alias_value":"NHE6LKZ2","created_at":"2026-07-05T05:57:28.798549+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.06624","citing_title":"Principles and Practice of Deep Representation Learning: or a Mathematical Theory of Memory","ref_index":118,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29358","citing_title":"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet","ref_index":76,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23040","citing_title":"Steered Generation via Gradient-Based Optimization on Sparse Query Features","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2309.08600","citing_title":"Sparse Autoencoders Find Highly Interpretable Features in Language Models","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2511.20233","citing_title":"REFLEX: Self-Refining Explainable Fact-Checking via Verdict-Anchored Style Control","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12813","citing_title":"REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinations","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2406.04093","citing_title":"Scaling and evaluating sparse autoencoders","ref_index":70,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NHE6LKZ2NNH66LELCBN3ZFUM6B","json":"https://pith.science/pith/NHE6LKZ2NNH66LELCBN3ZFUM6B.json","graph_json":"https://pith.science/api/pith-number/NHE6LKZ2NNH66LELCBN3ZFUM6B/graph.json","events_json":"https://pith.science/api/pith-number/NHE6LKZ2NNH66LELCBN3ZFUM6B/events.json","paper":"https://pith.science/paper/NHE6LKZ2"},"agent_actions":{"view_html":"https://pith.science/pith/NHE6LKZ2NNH66LELCBN3ZFUM6B","download_json":"https://pith.science/pith/NHE6LKZ2NNH66LELCBN3ZFUM6B.json","view_paper":"https://pith.science/paper/NHE6LKZ2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2103.15949&json=true","fetch_graph":"https://pith.science/api/pith-number/NHE6LKZ2NNH66LELCBN3ZFUM6B/graph.json","fetch_events":"https://pith.science/api/pith-number/NHE6LKZ2NNH66LELCBN3ZFUM6B/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NHE6LKZ2NNH66LELCBN3ZFUM6B/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NHE6LKZ2NNH66LELCBN3ZFUM6B/action/storage_attestation","attest_author":"https://pith.science/pith/NHE6LKZ2NNH66LELCBN3ZFUM6B/action/author_attestation","sign_citation":"https://pith.science/pith/NHE6LKZ2NNH66LELCBN3ZFUM6B/action/citation_signature","submit_replication":"https://pith.science/pith/NHE6LKZ2NNH66LELCBN3ZFUM6B/action/replication_record"}},"created_at":"2026-07-05T05:57:28.798549+00:00","updated_at":"2026-07-05T05:57:28.798549+00:00"}