{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:RGM6QJFNQCKUV42KGZJW34RL2J","short_pith_number":"pith:RGM6QJFN","schema_version":"1.0","canonical_sha256":"8999e824ad80954af34a36536df22bd25ea85b9c8c9009d045103d5293570c54","source":{"kind":"arxiv","id":"2203.13151","version":2},"attestation_state":"computed","paper":{"title":"Multi-armed bandits for resource efficient, online optimization of language model pre-training: the use case of dynamic masking","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.CL","authors_text":"I\\~nigo Urteaga, Moulay-Za\\\"idane Dra\\\"idia, Shahram Khadivi, Tomer Lancewicki","submitted_at":"2022-03-24T16:12:21Z","abstract_excerpt":"We design and evaluate a Bayesian optimization framework for resource efficient pre-training of Transformer-based language models (TLMs). TLM pre-training requires high computational resources and introduces many unresolved design choices, such as selecting its pre-training hyperparameters. We propose a multi-armed bandit framework for the sequential selection of TLM pre-training hyperparameters, aimed at optimizing language model performance, in a resource efficient manner. We design a Thompson sampling algorithm, with a surrogate Gaussian process reward model of the Masked Language Model (ML"},"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":"2203.13151","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-03-24T16:12:21Z","cross_cats_sorted":["cs.LG","stat.ML"],"title_canon_sha256":"8b4053d216862e25ff6e87669b60dbff4b27839cca76be43d4298ffada7de29d","abstract_canon_sha256":"9d782ba7190fd9e36b6040eba156302dc9570d5f612d38479a1b3ce8ffbb756d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:15:12.602527Z","signature_b64":"02gF6QsYEmfyQR9MyVLAlTZEr3b2tuminBLHHRzL4ZRRYIa81yYXV/k/SU91b4Dvhr3cQt5hSO0pgWEuviByCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8999e824ad80954af34a36536df22bd25ea85b9c8c9009d045103d5293570c54","last_reissued_at":"2026-07-05T06:15:12.602025Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:15:12.602025Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multi-armed bandits for resource efficient, online optimization of language model pre-training: the use case of dynamic masking","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.CL","authors_text":"I\\~nigo Urteaga, Moulay-Za\\\"idane Dra\\\"idia, Shahram Khadivi, Tomer Lancewicki","submitted_at":"2022-03-24T16:12:21Z","abstract_excerpt":"We design and evaluate a Bayesian optimization framework for resource efficient pre-training of Transformer-based language models (TLMs). TLM pre-training requires high computational resources and introduces many unresolved design choices, such as selecting its pre-training hyperparameters. We propose a multi-armed bandit framework for the sequential selection of TLM pre-training hyperparameters, aimed at optimizing language model performance, in a resource efficient manner. We design a Thompson sampling algorithm, with a surrogate Gaussian process reward model of the Masked Language Model (ML"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2203.13151","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/2203.13151/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":"2203.13151","created_at":"2026-07-05T06:15:12.602088+00:00"},{"alias_kind":"arxiv_version","alias_value":"2203.13151v2","created_at":"2026-07-05T06:15:12.602088+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2203.13151","created_at":"2026-07-05T06:15:12.602088+00:00"},{"alias_kind":"pith_short_12","alias_value":"RGM6QJFNQCKU","created_at":"2026-07-05T06:15:12.602088+00:00"},{"alias_kind":"pith_short_16","alias_value":"RGM6QJFNQCKUV42K","created_at":"2026-07-05T06:15:12.602088+00:00"},{"alias_kind":"pith_short_8","alias_value":"RGM6QJFN","created_at":"2026-07-05T06:15:12.602088+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RGM6QJFNQCKUV42KGZJW34RL2J","json":"https://pith.science/pith/RGM6QJFNQCKUV42KGZJW34RL2J.json","graph_json":"https://pith.science/api/pith-number/RGM6QJFNQCKUV42KGZJW34RL2J/graph.json","events_json":"https://pith.science/api/pith-number/RGM6QJFNQCKUV42KGZJW34RL2J/events.json","paper":"https://pith.science/paper/RGM6QJFN"},"agent_actions":{"view_html":"https://pith.science/pith/RGM6QJFNQCKUV42KGZJW34RL2J","download_json":"https://pith.science/pith/RGM6QJFNQCKUV42KGZJW34RL2J.json","view_paper":"https://pith.science/paper/RGM6QJFN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2203.13151&json=true","fetch_graph":"https://pith.science/api/pith-number/RGM6QJFNQCKUV42KGZJW34RL2J/graph.json","fetch_events":"https://pith.science/api/pith-number/RGM6QJFNQCKUV42KGZJW34RL2J/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RGM6QJFNQCKUV42KGZJW34RL2J/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RGM6QJFNQCKUV42KGZJW34RL2J/action/storage_attestation","attest_author":"https://pith.science/pith/RGM6QJFNQCKUV42KGZJW34RL2J/action/author_attestation","sign_citation":"https://pith.science/pith/RGM6QJFNQCKUV42KGZJW34RL2J/action/citation_signature","submit_replication":"https://pith.science/pith/RGM6QJFNQCKUV42KGZJW34RL2J/action/replication_record"}},"created_at":"2026-07-05T06:15:12.602088+00:00","updated_at":"2026-07-05T06:15:12.602088+00:00"}