{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:AUV4XHSYFCR5GRH7XV6JV6CGX2","short_pith_number":"pith:AUV4XHSY","schema_version":"1.0","canonical_sha256":"052bcb9e5828a3d344ffbd7c9af846bea8a93ae2b149cdee254cb6d84194dd8e","source":{"kind":"arxiv","id":"2410.06833","version":1},"attestation_state":"computed","paper":{"title":"Dynamic metastability in the self-attention model","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.AP","math.DS"],"primary_cat":"cs.LG","authors_text":"Borjan Geshkovski, Hugo Koubbi, Philippe Rigollet, Yury Polyanskiy","submitted_at":"2024-10-09T12:50:50Z","abstract_excerpt":"We consider the self-attention model - an interacting particle system on the unit sphere, which serves as a toy model for Transformers, the deep neural network architecture behind the recent successes of large language models. We prove the appearance of dynamic metastability conjectured in [GLPR23] - although particles collapse to a single cluster in infinite time, they remain trapped near a configuration of several clusters for an exponentially long period of time. By leveraging a gradient flow interpretation of the system, we also connect our result to an overarching framework of slow motion"},"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":"2410.06833","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-10-09T12:50:50Z","cross_cats_sorted":["math.AP","math.DS"],"title_canon_sha256":"8533e300ae24377ea6317298297a3a5253422b93d9c74caa98e58991e7f9380c","abstract_canon_sha256":"ce07645cf5bb52aa9c5fb876384a0934591aa7fee1284184979b256d2331f216"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:18:07.492652Z","signature_b64":"lO2/LHSAycCwUgs1F1MDSHmLr9KkdAkGTgs+XxfRrfOnmWHY1Dxb/tNxOsgSHJk57jtuDzSNjpNo0j5mRmLzBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"052bcb9e5828a3d344ffbd7c9af846bea8a93ae2b149cdee254cb6d84194dd8e","last_reissued_at":"2026-07-05T09:18:07.492165Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:18:07.492165Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dynamic metastability in the self-attention model","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.AP","math.DS"],"primary_cat":"cs.LG","authors_text":"Borjan Geshkovski, Hugo Koubbi, Philippe Rigollet, Yury Polyanskiy","submitted_at":"2024-10-09T12:50:50Z","abstract_excerpt":"We consider the self-attention model - an interacting particle system on the unit sphere, which serves as a toy model for Transformers, the deep neural network architecture behind the recent successes of large language models. We prove the appearance of dynamic metastability conjectured in [GLPR23] - although particles collapse to a single cluster in infinite time, they remain trapped near a configuration of several clusters for an exponentially long period of time. By leveraging a gradient flow interpretation of the system, we also connect our result to an overarching framework of slow motion"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.06833","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/2410.06833/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":"2410.06833","created_at":"2026-07-05T09:18:07.492221+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.06833v1","created_at":"2026-07-05T09:18:07.492221+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.06833","created_at":"2026-07-05T09:18:07.492221+00:00"},{"alias_kind":"pith_short_12","alias_value":"AUV4XHSYFCR5","created_at":"2026-07-05T09:18:07.492221+00:00"},{"alias_kind":"pith_short_16","alias_value":"AUV4XHSYFCR5GRH7","created_at":"2026-07-05T09:18:07.492221+00:00"},{"alias_kind":"pith_short_8","alias_value":"AUV4XHSY","created_at":"2026-07-05T09:18:07.492221+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.25619","citing_title":"Analogies between Transformer Layers and Power Method","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07600","citing_title":"Reachability and asymptotics of Gaussian Transformer dynamics","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2504.14697","citing_title":"Quantitative Clustering in Mean-Field Transformer Models","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16747","citing_title":"Propagation of Chaos in Contextual Flow Maps","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12697","citing_title":"A Unified Framework for Critical Scaling of Inverse Temperature in Self-Attention","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26898","citing_title":"Stochastic Scaling Limits and Synchronization by Noise in Deep Transformer Models","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09213","citing_title":"Kinetic theory for Transformers and the lost-in-the-middle phenomenon","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10931","citing_title":"Quantifying Concentration Phenomena of Mean-Field Transformers in the Low-Temperature Regime","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26085","citing_title":"Spectral Selection in Symmetric Self-Attention Dynamics","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04279","citing_title":"Gradient Flow Structure and Quantitative Dynamics of Multi-Head Self-Attention","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04279","citing_title":"Gradient Flow Structure and Quantitative Dynamics of Multi-Head Self-Attention","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AUV4XHSYFCR5GRH7XV6JV6CGX2","json":"https://pith.science/pith/AUV4XHSYFCR5GRH7XV6JV6CGX2.json","graph_json":"https://pith.science/api/pith-number/AUV4XHSYFCR5GRH7XV6JV6CGX2/graph.json","events_json":"https://pith.science/api/pith-number/AUV4XHSYFCR5GRH7XV6JV6CGX2/events.json","paper":"https://pith.science/paper/AUV4XHSY"},"agent_actions":{"view_html":"https://pith.science/pith/AUV4XHSYFCR5GRH7XV6JV6CGX2","download_json":"https://pith.science/pith/AUV4XHSYFCR5GRH7XV6JV6CGX2.json","view_paper":"https://pith.science/paper/AUV4XHSY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.06833&json=true","fetch_graph":"https://pith.science/api/pith-number/AUV4XHSYFCR5GRH7XV6JV6CGX2/graph.json","fetch_events":"https://pith.science/api/pith-number/AUV4XHSYFCR5GRH7XV6JV6CGX2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AUV4XHSYFCR5GRH7XV6JV6CGX2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AUV4XHSYFCR5GRH7XV6JV6CGX2/action/storage_attestation","attest_author":"https://pith.science/pith/AUV4XHSYFCR5GRH7XV6JV6CGX2/action/author_attestation","sign_citation":"https://pith.science/pith/AUV4XHSYFCR5GRH7XV6JV6CGX2/action/citation_signature","submit_replication":"https://pith.science/pith/AUV4XHSYFCR5GRH7XV6JV6CGX2/action/replication_record"}},"created_at":"2026-07-05T09:18:07.492221+00:00","updated_at":"2026-07-05T09:18:07.492221+00:00"}