{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:G22N2X3KENBGQCSEF3B3IMJHE3","short_pith_number":"pith:G22N2X3K","schema_version":"1.0","canonical_sha256":"36b4dd5f6a2342680a442ec3b4312726d964f5a03e8b725e5a1b04d75157dbdf","source":{"kind":"arxiv","id":"2408.15138","version":3},"attestation_state":"computed","paper":{"title":"How transformers learn structured data: insights from hierarchical filtering","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cond-mat.dis-nn","cond-mat.stat-mech","cs.CL"],"primary_cat":"cs.LG","authors_text":"Emanuele Moscato, Jerome Garnier-Brun, Luca Saglietti, Marc M\\'ezard","submitted_at":"2024-08-27T15:23:09Z","abstract_excerpt":"Understanding the learning process and the embedded computation in transformers is becoming a central goal for the development of interpretable AI. In the present study, we introduce a hierarchical filtering procedure for data models of sequences on trees, allowing us to hand-tune the range of positional correlations in the data. Leveraging this controlled setting, we provide evidence that vanilla encoder-only transformers can approximate the exact inference algorithm when trained on root classification and masked language modeling tasks, and study how this computation is discovered and implem"},"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":"2408.15138","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-08-27T15:23:09Z","cross_cats_sorted":["cond-mat.dis-nn","cond-mat.stat-mech","cs.CL"],"title_canon_sha256":"f486b29307b15b8ccf35e84d055a39cc504f57cc5b392918fae86cae2e8b49cf","abstract_canon_sha256":"c4a068a3dfba61dd50d39107df58db5591880c37fa8cd2b9e2494e83e4d4ecc2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:19:00.267170Z","signature_b64":"3DkmRCvgb48sNZJCTO+xFU1APkoWjhAdIdPYZPhQIxbotBgWtN3YOc19oeh+oJtsv5pQxC68uK2T8+lWIgEXCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"36b4dd5f6a2342680a442ec3b4312726d964f5a03e8b725e5a1b04d75157dbdf","last_reissued_at":"2026-07-05T11:19:00.266675Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:19:00.266675Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How transformers learn structured data: insights from hierarchical filtering","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cond-mat.dis-nn","cond-mat.stat-mech","cs.CL"],"primary_cat":"cs.LG","authors_text":"Emanuele Moscato, Jerome Garnier-Brun, Luca Saglietti, Marc M\\'ezard","submitted_at":"2024-08-27T15:23:09Z","abstract_excerpt":"Understanding the learning process and the embedded computation in transformers is becoming a central goal for the development of interpretable AI. In the present study, we introduce a hierarchical filtering procedure for data models of sequences on trees, allowing us to hand-tune the range of positional correlations in the data. Leveraging this controlled setting, we provide evidence that vanilla encoder-only transformers can approximate the exact inference algorithm when trained on root classification and masked language modeling tasks, and study how this computation is discovered and implem"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.15138","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/2408.15138/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":"2408.15138","created_at":"2026-07-05T11:19:00.266737+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.15138v3","created_at":"2026-07-05T11:19:00.266737+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.15138","created_at":"2026-07-05T11:19:00.266737+00:00"},{"alias_kind":"pith_short_12","alias_value":"G22N2X3KENBG","created_at":"2026-07-05T11:19:00.266737+00:00"},{"alias_kind":"pith_short_16","alias_value":"G22N2X3KENBGQCSE","created_at":"2026-07-05T11:19:00.266737+00:00"},{"alias_kind":"pith_short_8","alias_value":"G22N2X3K","created_at":"2026-07-05T11:19:00.266737+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.20299","citing_title":"Statistical Properties of Training & Generalization","ref_index":212,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20347","citing_title":"Critical Percolation as a Synthetic Data Model for Interpretability","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20299","citing_title":"Statistical Properties of Training & Generalization","ref_index":212,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28103","citing_title":"Phase structure of the Random Language Model","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27006","citing_title":"Sampling Data with Chains of Forward-Backward Diffusion Steps","ref_index":57,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27734","citing_title":"Learn from your own latents and not from tokens: A sample-complexity theory","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07844","citing_title":"Distributional simplicity bias and effective convexity in Energy Based Models","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/G22N2X3KENBGQCSEF3B3IMJHE3","json":"https://pith.science/pith/G22N2X3KENBGQCSEF3B3IMJHE3.json","graph_json":"https://pith.science/api/pith-number/G22N2X3KENBGQCSEF3B3IMJHE3/graph.json","events_json":"https://pith.science/api/pith-number/G22N2X3KENBGQCSEF3B3IMJHE3/events.json","paper":"https://pith.science/paper/G22N2X3K"},"agent_actions":{"view_html":"https://pith.science/pith/G22N2X3KENBGQCSEF3B3IMJHE3","download_json":"https://pith.science/pith/G22N2X3KENBGQCSEF3B3IMJHE3.json","view_paper":"https://pith.science/paper/G22N2X3K","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.15138&json=true","fetch_graph":"https://pith.science/api/pith-number/G22N2X3KENBGQCSEF3B3IMJHE3/graph.json","fetch_events":"https://pith.science/api/pith-number/G22N2X3KENBGQCSEF3B3IMJHE3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/G22N2X3KENBGQCSEF3B3IMJHE3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/G22N2X3KENBGQCSEF3B3IMJHE3/action/storage_attestation","attest_author":"https://pith.science/pith/G22N2X3KENBGQCSEF3B3IMJHE3/action/author_attestation","sign_citation":"https://pith.science/pith/G22N2X3KENBGQCSEF3B3IMJHE3/action/citation_signature","submit_replication":"https://pith.science/pith/G22N2X3KENBGQCSEF3B3IMJHE3/action/replication_record"}},"created_at":"2026-07-05T11:19:00.266737+00:00","updated_at":"2026-07-05T11:19:00.266737+00:00"}