{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:WRT7ELWAHT2I7X2Y7AZHTBPMHT","short_pith_number":"pith:WRT7ELWA","schema_version":"1.0","canonical_sha256":"b467f22ec03cf48fdf58f8327985ec3ce03d1778a6a8152b68cc04cf297e5709","source":{"kind":"arxiv","id":"2508.17821","version":3},"attestation_state":"computed","paper":{"title":"Limitations of Normalization in Attention Mechanism","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Mikhail Burtsev, Radu State, Tatiana Petrova, Timur Mudarisov","submitted_at":"2025-08-25T09:25:05Z","abstract_excerpt":"This paper investigates the limitations of the normalization in attention mechanisms. We begin with a theoretical framework that enables the identification of the model's selective ability and the geometric separation involved in token selection. Our analysis includes explicit bounds on distances and separation criteria for token vectors under softmax scaling. Through experiments with pre-trained GPT-2 model, we empirically validate our theoretical results and analyze key behaviors of the attention mechanism. Notably, we demonstrate that as the number of selected tokens increases, the model's "},"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":"2508.17821","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-25T09:25:05Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"87831402b9bbe659a668b6d791661718242521e8936b2c1cb8ed71d5b6633211","abstract_canon_sha256":"1a478398aaea1bed8e41126a3ec4f12e51db41855bc9d3fe3e6e9e066955e50c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-08T01:05:02.483709Z","signature_b64":"wrZagLcD7DGpeu6ZsY4Q3FHXrcrqWaZVaPDl5Ysawl//yW7IW4iHvKyAaFlCSE1vK5XUg2xY70PiMjCwXw3uDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b467f22ec03cf48fdf58f8327985ec3ce03d1778a6a8152b68cc04cf297e5709","last_reissued_at":"2026-06-08T01:05:02.482804Z","signature_status":"signed_v1","first_computed_at":"2026-06-08T01:05:02.482804Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Limitations of Normalization in Attention Mechanism","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Mikhail Burtsev, Radu State, Tatiana Petrova, Timur Mudarisov","submitted_at":"2025-08-25T09:25:05Z","abstract_excerpt":"This paper investigates the limitations of the normalization in attention mechanisms. We begin with a theoretical framework that enables the identification of the model's selective ability and the geometric separation involved in token selection. Our analysis includes explicit bounds on distances and separation criteria for token vectors under softmax scaling. Through experiments with pre-trained GPT-2 model, we empirically validate our theoretical results and analyze key behaviors of the attention mechanism. Notably, we demonstrate that as the number of selected tokens increases, the model's "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.17821","kind":"arxiv","version":3},"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/2508.17821/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":"2508.17821","created_at":"2026-06-08T01:05:02.482928+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.17821v3","created_at":"2026-06-08T01:05:02.482928+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.17821","created_at":"2026-06-08T01:05:02.482928+00:00"},{"alias_kind":"pith_short_12","alias_value":"WRT7ELWAHT2I","created_at":"2026-06-08T01:05:02.482928+00:00"},{"alias_kind":"pith_short_16","alias_value":"WRT7ELWAHT2I7X2Y","created_at":"2026-06-08T01:05:02.482928+00:00"},{"alias_kind":"pith_short_8","alias_value":"WRT7ELWA","created_at":"2026-06-08T01:05:02.482928+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":5,"sample":[{"citing_arxiv_id":"2605.20275","citing_title":"You Don't Need Attention: Gated Convolutional Modeling for Watch-Based Fall Detection","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2604.28161","citing_title":"RopeDreamer: A Kinematic Recurrent State Space Model for Dynamics of Flexible Deformable Linear Objects","ref_index":24,"is_internal_anchor":true},{"citing_arxiv_id":"2604.21570","citing_title":"SpecSyn: LLM-based Synthesis and Refinement of Formal Specifications for Real-world Program Verification","ref_index":56,"is_internal_anchor":true},{"citing_arxiv_id":"2604.11182","citing_title":"Evaluating Memory Capability in Continuous Lifelog Scenario","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2604.18580","citing_title":"Sessa: Selective State Space Attention","ref_index":3,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WRT7ELWAHT2I7X2Y7AZHTBPMHT","json":"https://pith.science/pith/WRT7ELWAHT2I7X2Y7AZHTBPMHT.json","graph_json":"https://pith.science/api/pith-number/WRT7ELWAHT2I7X2Y7AZHTBPMHT/graph.json","events_json":"https://pith.science/api/pith-number/WRT7ELWAHT2I7X2Y7AZHTBPMHT/events.json","paper":"https://pith.science/paper/WRT7ELWA"},"agent_actions":{"view_html":"https://pith.science/pith/WRT7ELWAHT2I7X2Y7AZHTBPMHT","download_json":"https://pith.science/pith/WRT7ELWAHT2I7X2Y7AZHTBPMHT.json","view_paper":"https://pith.science/paper/WRT7ELWA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.17821&json=true","fetch_graph":"https://pith.science/api/pith-number/WRT7ELWAHT2I7X2Y7AZHTBPMHT/graph.json","fetch_events":"https://pith.science/api/pith-number/WRT7ELWAHT2I7X2Y7AZHTBPMHT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WRT7ELWAHT2I7X2Y7AZHTBPMHT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WRT7ELWAHT2I7X2Y7AZHTBPMHT/action/storage_attestation","attest_author":"https://pith.science/pith/WRT7ELWAHT2I7X2Y7AZHTBPMHT/action/author_attestation","sign_citation":"https://pith.science/pith/WRT7ELWAHT2I7X2Y7AZHTBPMHT/action/citation_signature","submit_replication":"https://pith.science/pith/WRT7ELWAHT2I7X2Y7AZHTBPMHT/action/replication_record"}},"created_at":"2026-06-08T01:05:02.482928+00:00","updated_at":"2026-06-08T01:05:02.482928+00:00"}