{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ML5WIV56P7VMM3UJXTICVJ23JW","short_pith_number":"pith:ML5WIV56","schema_version":"1.0","canonical_sha256":"62fb6457be7feac66e89bcd02aa75b4da2ee5d00b89b0758d86e649afa0eac76","source":{"kind":"arxiv","id":"2505.05648","version":1},"attestation_state":"computed","paper":{"title":"Privacy-Preserving Transformers: SwiftKey's Differential Privacy Implementation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR","cs.LG"],"primary_cat":"cs.CL","authors_text":"Abdelrahman Abouelenin, Amr Hendy, Mohamed Abdelrehim, Mohamed Afify, Raffy Fahim","submitted_at":"2025-05-08T21:08:04Z","abstract_excerpt":"In this paper we train a transformer using differential privacy (DP) for language modeling in SwiftKey. We run multiple experiments to balance the trade-off between the model size, run-time speed and accuracy. We show that we get small and consistent gains in the next-word-prediction and accuracy with graceful increase in memory and speed compared to the production GRU. This is obtained by scaling down a GPT2 architecture to fit the required size and a two stage training process that builds a seed model on general data and DP finetunes it on typing data. The transformer is integrated using ONN"},"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":"2505.05648","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-08T21:08:04Z","cross_cats_sorted":["cs.CR","cs.LG"],"title_canon_sha256":"8c5fbe1951b83dbe631389ec75b834b06dcebdfa56d47048fc9ae6c09a70c61d","abstract_canon_sha256":"750241af61c48d3567c9aaedba44536456bf19c0e9490f7bb422e4b46f1d1a01"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:00:45.293835Z","signature_b64":"FuOpq/3K4YT83wNWC3Sni4AsrKBcoFwOITzYVHthiNfmj+cu/8ZDYyXZukDM6ux9wFgOCY8sHcwwgzlamHpHBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"62fb6457be7feac66e89bcd02aa75b4da2ee5d00b89b0758d86e649afa0eac76","last_reissued_at":"2026-07-05T11:00:45.293388Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:00:45.293388Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Privacy-Preserving Transformers: SwiftKey's Differential Privacy Implementation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR","cs.LG"],"primary_cat":"cs.CL","authors_text":"Abdelrahman Abouelenin, Amr Hendy, Mohamed Abdelrehim, Mohamed Afify, Raffy Fahim","submitted_at":"2025-05-08T21:08:04Z","abstract_excerpt":"In this paper we train a transformer using differential privacy (DP) for language modeling in SwiftKey. We run multiple experiments to balance the trade-off between the model size, run-time speed and accuracy. We show that we get small and consistent gains in the next-word-prediction and accuracy with graceful increase in memory and speed compared to the production GRU. This is obtained by scaling down a GPT2 architecture to fit the required size and a two stage training process that builds a seed model on general data and DP finetunes it on typing data. The transformer is integrated using ONN"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.05648","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/2505.05648/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":"2505.05648","created_at":"2026-07-05T11:00:45.293441+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.05648v1","created_at":"2026-07-05T11:00:45.293441+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.05648","created_at":"2026-07-05T11:00:45.293441+00:00"},{"alias_kind":"pith_short_12","alias_value":"ML5WIV56P7VM","created_at":"2026-07-05T11:00:45.293441+00:00"},{"alias_kind":"pith_short_16","alias_value":"ML5WIV56P7VMM3UJ","created_at":"2026-07-05T11:00:45.293441+00:00"},{"alias_kind":"pith_short_8","alias_value":"ML5WIV56","created_at":"2026-07-05T11:00:45.293441+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/ML5WIV56P7VMM3UJXTICVJ23JW","json":"https://pith.science/pith/ML5WIV56P7VMM3UJXTICVJ23JW.json","graph_json":"https://pith.science/api/pith-number/ML5WIV56P7VMM3UJXTICVJ23JW/graph.json","events_json":"https://pith.science/api/pith-number/ML5WIV56P7VMM3UJXTICVJ23JW/events.json","paper":"https://pith.science/paper/ML5WIV56"},"agent_actions":{"view_html":"https://pith.science/pith/ML5WIV56P7VMM3UJXTICVJ23JW","download_json":"https://pith.science/pith/ML5WIV56P7VMM3UJXTICVJ23JW.json","view_paper":"https://pith.science/paper/ML5WIV56","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.05648&json=true","fetch_graph":"https://pith.science/api/pith-number/ML5WIV56P7VMM3UJXTICVJ23JW/graph.json","fetch_events":"https://pith.science/api/pith-number/ML5WIV56P7VMM3UJXTICVJ23JW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ML5WIV56P7VMM3UJXTICVJ23JW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ML5WIV56P7VMM3UJXTICVJ23JW/action/storage_attestation","attest_author":"https://pith.science/pith/ML5WIV56P7VMM3UJXTICVJ23JW/action/author_attestation","sign_citation":"https://pith.science/pith/ML5WIV56P7VMM3UJXTICVJ23JW/action/citation_signature","submit_replication":"https://pith.science/pith/ML5WIV56P7VMM3UJXTICVJ23JW/action/replication_record"}},"created_at":"2026-07-05T11:00:45.293441+00:00","updated_at":"2026-07-05T11:00:45.293441+00:00"}