{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:FC3N34R6SUP3NTEO4CNG76IELV","short_pith_number":"pith:FC3N34R6","schema_version":"1.0","canonical_sha256":"28b6ddf23e951fb6cc8ee09a6ff9045d63bad4ae31e7c7a2e51437343f3be5b1","source":{"kind":"arxiv","id":"2401.12143","version":2},"attestation_state":"computed","paper":{"title":"Anisotropy Is Inherent to Self-Attention in Transformers","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Beno\\^it Sagot, \\'Eric de la Clergerie, Nathan Godey","submitted_at":"2024-01-22T17:26:55Z","abstract_excerpt":"The representation degeneration problem is a phenomenon that is widely observed among self-supervised learning methods based on Transformers. In NLP, it takes the form of anisotropy, a singular property of hidden representations which makes them unexpectedly close to each other in terms of angular distance (cosine-similarity). Some recent works tend to show that anisotropy is a consequence of optimizing the cross-entropy loss on long-tailed distributions of tokens. We show in this paper that anisotropy can also be observed empirically in language models with specific objectives that should not"},"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":"2401.12143","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-01-22T17:26:55Z","cross_cats_sorted":[],"title_canon_sha256":"96141d67a20594982c8738b70c3f872eea02cb9953e1c9e674f10a900387d233","abstract_canon_sha256":"b816fc333cd92d33ebbfd5c3287a1d13f0f1a732b122de97bcc8fc6b62459d2b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:36:58.176378Z","signature_b64":"WVaSxj/cvjtNK7iF/d3MDrWcPqq7UtH0qAOZgs7ML5RQDW4n8TWxwiI+/Zq2vuZAlNcjEDQudq/D7R2Mq2fNAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"28b6ddf23e951fb6cc8ee09a6ff9045d63bad4ae31e7c7a2e51437343f3be5b1","last_reissued_at":"2026-07-05T07:36:58.175945Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:36:58.175945Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Anisotropy Is Inherent to Self-Attention in Transformers","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Beno\\^it Sagot, \\'Eric de la Clergerie, Nathan Godey","submitted_at":"2024-01-22T17:26:55Z","abstract_excerpt":"The representation degeneration problem is a phenomenon that is widely observed among self-supervised learning methods based on Transformers. In NLP, it takes the form of anisotropy, a singular property of hidden representations which makes them unexpectedly close to each other in terms of angular distance (cosine-similarity). Some recent works tend to show that anisotropy is a consequence of optimizing the cross-entropy loss on long-tailed distributions of tokens. We show in this paper that anisotropy can also be observed empirically in language models with specific objectives that should not"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.12143","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/2401.12143/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":"2401.12143","created_at":"2026-07-05T07:36:58.176001+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.12143v2","created_at":"2026-07-05T07:36:58.176001+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.12143","created_at":"2026-07-05T07:36:58.176001+00:00"},{"alias_kind":"pith_short_12","alias_value":"FC3N34R6SUP3","created_at":"2026-07-05T07:36:58.176001+00:00"},{"alias_kind":"pith_short_16","alias_value":"FC3N34R6SUP3NTEO","created_at":"2026-07-05T07:36:58.176001+00:00"},{"alias_kind":"pith_short_8","alias_value":"FC3N34R6","created_at":"2026-07-05T07:36:58.176001+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.27378","citing_title":"Formalizing Latent Thoughts: Four Axioms of Thought Representation in LLMs","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2510.24561","citing_title":"LoRA-DA: Data-Aware Initialization for Low-Rank Adaptation via Asymptotic Analysis","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FC3N34R6SUP3NTEO4CNG76IELV","json":"https://pith.science/pith/FC3N34R6SUP3NTEO4CNG76IELV.json","graph_json":"https://pith.science/api/pith-number/FC3N34R6SUP3NTEO4CNG76IELV/graph.json","events_json":"https://pith.science/api/pith-number/FC3N34R6SUP3NTEO4CNG76IELV/events.json","paper":"https://pith.science/paper/FC3N34R6"},"agent_actions":{"view_html":"https://pith.science/pith/FC3N34R6SUP3NTEO4CNG76IELV","download_json":"https://pith.science/pith/FC3N34R6SUP3NTEO4CNG76IELV.json","view_paper":"https://pith.science/paper/FC3N34R6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.12143&json=true","fetch_graph":"https://pith.science/api/pith-number/FC3N34R6SUP3NTEO4CNG76IELV/graph.json","fetch_events":"https://pith.science/api/pith-number/FC3N34R6SUP3NTEO4CNG76IELV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FC3N34R6SUP3NTEO4CNG76IELV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FC3N34R6SUP3NTEO4CNG76IELV/action/storage_attestation","attest_author":"https://pith.science/pith/FC3N34R6SUP3NTEO4CNG76IELV/action/author_attestation","sign_citation":"https://pith.science/pith/FC3N34R6SUP3NTEO4CNG76IELV/action/citation_signature","submit_replication":"https://pith.science/pith/FC3N34R6SUP3NTEO4CNG76IELV/action/replication_record"}},"created_at":"2026-07-05T07:36:58.176001+00:00","updated_at":"2026-07-05T07:36:58.176001+00:00"}