{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:UUMEE3WLVHGU6YREXI6PPR2C54","short_pith_number":"pith:UUMEE3WL","schema_version":"1.0","canonical_sha256":"a518426ecba9cd4f6224ba3cf7c742ef358facc09a9ecab982fac7f5ce45db09","source":{"kind":"arxiv","id":"2402.04161","version":2},"attestation_state":"computed","paper":{"title":"Attention with Markov: A Framework for Principled Analysis of Transformers via Markov Chains","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.IT","math.IT","stat.ML"],"primary_cat":"cs.LG","authors_text":"Adway Girish, Alliot Nagle, Ashok Vardhan Makkuva, Hyeji Kim, Marco Bondaschi, Martin Jaggi, Michael Gastpar","submitted_at":"2024-02-06T17:18:59Z","abstract_excerpt":"Attention-based transformers have achieved tremendous success across a variety of disciplines including natural languages. To deepen our understanding of their sequential modeling capabilities, there is a growing interest in using Markov input processes to study them. A key finding is that when trained on first-order Markov chains, transformers with two or more layers consistently develop an induction head mechanism to estimate the in-context bigram conditional distribution. In contrast, single-layer transformers, unable to form an induction head, directly learn the Markov kernel but often fac"},"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":"2402.04161","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-06T17:18:59Z","cross_cats_sorted":["cs.CL","cs.IT","math.IT","stat.ML"],"title_canon_sha256":"2fe7ee210554f24a15bf8392f212298e1ce677fb9db24105381775bb52cdf3eb","abstract_canon_sha256":"dba9b15193d85389810295aeb7b7ea3c143c9e63334d55cb05e1866588c7aa20"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:40:29.850766Z","signature_b64":"s2sK9C7OkBGoLFYci3p5xoWc3FfYD1TW+gRAz9lmPz4fVDfMsepzykWqWsg6gto4fq7mEdk5VBIXTyNAVxogDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a518426ecba9cd4f6224ba3cf7c742ef358facc09a9ecab982fac7f5ce45db09","last_reissued_at":"2026-07-05T11:40:29.850204Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:40:29.850204Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Attention with Markov: A Framework for Principled Analysis of Transformers via Markov Chains","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.IT","math.IT","stat.ML"],"primary_cat":"cs.LG","authors_text":"Adway Girish, Alliot Nagle, Ashok Vardhan Makkuva, Hyeji Kim, Marco Bondaschi, Martin Jaggi, Michael Gastpar","submitted_at":"2024-02-06T17:18:59Z","abstract_excerpt":"Attention-based transformers have achieved tremendous success across a variety of disciplines including natural languages. To deepen our understanding of their sequential modeling capabilities, there is a growing interest in using Markov input processes to study them. A key finding is that when trained on first-order Markov chains, transformers with two or more layers consistently develop an induction head mechanism to estimate the in-context bigram conditional distribution. In contrast, single-layer transformers, unable to form an induction head, directly learn the Markov kernel but often fac"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.04161","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/2402.04161/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":"2402.04161","created_at":"2026-07-05T11:40:29.850264+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.04161v2","created_at":"2026-07-05T11:40:29.850264+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.04161","created_at":"2026-07-05T11:40:29.850264+00:00"},{"alias_kind":"pith_short_12","alias_value":"UUMEE3WLVHGU","created_at":"2026-07-05T11:40:29.850264+00:00"},{"alias_kind":"pith_short_16","alias_value":"UUMEE3WLVHGU6YRE","created_at":"2026-07-05T11:40:29.850264+00:00"},{"alias_kind":"pith_short_8","alias_value":"UUMEE3WL","created_at":"2026-07-05T11:40:29.850264+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26749","citing_title":"Structure Before Collapse: Transient semantic geometry in next-token prediction","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2503.10574","citing_title":"The LZ78 Source","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2506.07298","citing_title":"Pre-trained Large Language Models Learn Hidden Markov Models In-context","ref_index":34,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UUMEE3WLVHGU6YREXI6PPR2C54","json":"https://pith.science/pith/UUMEE3WLVHGU6YREXI6PPR2C54.json","graph_json":"https://pith.science/api/pith-number/UUMEE3WLVHGU6YREXI6PPR2C54/graph.json","events_json":"https://pith.science/api/pith-number/UUMEE3WLVHGU6YREXI6PPR2C54/events.json","paper":"https://pith.science/paper/UUMEE3WL"},"agent_actions":{"view_html":"https://pith.science/pith/UUMEE3WLVHGU6YREXI6PPR2C54","download_json":"https://pith.science/pith/UUMEE3WLVHGU6YREXI6PPR2C54.json","view_paper":"https://pith.science/paper/UUMEE3WL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.04161&json=true","fetch_graph":"https://pith.science/api/pith-number/UUMEE3WLVHGU6YREXI6PPR2C54/graph.json","fetch_events":"https://pith.science/api/pith-number/UUMEE3WLVHGU6YREXI6PPR2C54/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UUMEE3WLVHGU6YREXI6PPR2C54/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UUMEE3WLVHGU6YREXI6PPR2C54/action/storage_attestation","attest_author":"https://pith.science/pith/UUMEE3WLVHGU6YREXI6PPR2C54/action/author_attestation","sign_citation":"https://pith.science/pith/UUMEE3WLVHGU6YREXI6PPR2C54/action/citation_signature","submit_replication":"https://pith.science/pith/UUMEE3WLVHGU6YREXI6PPR2C54/action/replication_record"}},"created_at":"2026-07-05T11:40:29.850264+00:00","updated_at":"2026-07-05T11:40:29.850264+00:00"}