{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:IO4FOQTH7KJU5IGQCFMAT4ZV46","short_pith_number":"pith:IO4FOQTH","schema_version":"1.0","canonical_sha256":"43b8574267fa934ea0d0115809f335e795336f868f84bc031781af661d074a49","source":{"kind":"arxiv","id":"2306.07656","version":1},"attestation_state":"computed","paper":{"title":"Is Anisotropy Inherent to 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":"2023-06-13T09:54:01Z","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":"2306.07656","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2023-06-13T09:54:01Z","cross_cats_sorted":[],"title_canon_sha256":"b042e4ab4351aeba30fc6e30de3d812715b8e3f173faf3e3ead7f408c96f6ee4","abstract_canon_sha256":"895da2c83f04fabb178b1029ac1473f0b24a5b866e58be72c785c7d15640e89f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:20:07.345010Z","signature_b64":"nlf4VKouhV5jqTn/6BarbkvEGoBbXEEF6Dn0odXq+JVzuk3deUg5mQzccRqUxQjXzhr9pzctHaDUBhv51KJCAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"43b8574267fa934ea0d0115809f335e795336f868f84bc031781af661d074a49","last_reissued_at":"2026-07-05T06:20:07.344613Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:20:07.344613Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Is Anisotropy Inherent to 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":"2023-06-13T09:54:01Z","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":"2306.07656","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/2306.07656/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":"2306.07656","created_at":"2026-07-05T06:20:07.344674+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.07656v1","created_at":"2026-07-05T06:20:07.344674+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.07656","created_at":"2026-07-05T06:20:07.344674+00:00"},{"alias_kind":"pith_short_12","alias_value":"IO4FOQTH7KJU","created_at":"2026-07-05T06:20:07.344674+00:00"},{"alias_kind":"pith_short_16","alias_value":"IO4FOQTH7KJU5IGQ","created_at":"2026-07-05T06:20:07.344674+00:00"},{"alias_kind":"pith_short_8","alias_value":"IO4FOQTH","created_at":"2026-07-05T06:20:07.344674+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2411.14073","citing_title":"Meaning at the Planck scale? Contextualized word embeddings for doing history, philosophy, and sociology of science","ref_index":23,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IO4FOQTH7KJU5IGQCFMAT4ZV46","json":"https://pith.science/pith/IO4FOQTH7KJU5IGQCFMAT4ZV46.json","graph_json":"https://pith.science/api/pith-number/IO4FOQTH7KJU5IGQCFMAT4ZV46/graph.json","events_json":"https://pith.science/api/pith-number/IO4FOQTH7KJU5IGQCFMAT4ZV46/events.json","paper":"https://pith.science/paper/IO4FOQTH"},"agent_actions":{"view_html":"https://pith.science/pith/IO4FOQTH7KJU5IGQCFMAT4ZV46","download_json":"https://pith.science/pith/IO4FOQTH7KJU5IGQCFMAT4ZV46.json","view_paper":"https://pith.science/paper/IO4FOQTH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.07656&json=true","fetch_graph":"https://pith.science/api/pith-number/IO4FOQTH7KJU5IGQCFMAT4ZV46/graph.json","fetch_events":"https://pith.science/api/pith-number/IO4FOQTH7KJU5IGQCFMAT4ZV46/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IO4FOQTH7KJU5IGQCFMAT4ZV46/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IO4FOQTH7KJU5IGQCFMAT4ZV46/action/storage_attestation","attest_author":"https://pith.science/pith/IO4FOQTH7KJU5IGQCFMAT4ZV46/action/author_attestation","sign_citation":"https://pith.science/pith/IO4FOQTH7KJU5IGQCFMAT4ZV46/action/citation_signature","submit_replication":"https://pith.science/pith/IO4FOQTH7KJU5IGQCFMAT4ZV46/action/replication_record"}},"created_at":"2026-07-05T06:20:07.344674+00:00","updated_at":"2026-07-05T06:20:07.344674+00:00"}