{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:UDMVTLMYCCZNG33W6QQM6Z5T44","short_pith_number":"pith:UDMVTLMY","schema_version":"1.0","canonical_sha256":"a0d959ad9810b2d36f76f420cf67b3e722668cf65005c5232bf74f88ade52376","source":{"kind":"arxiv","id":"2010.05874","version":1},"attestation_state":"computed","paper":{"title":"Gradient Vaccine: Investigating and Improving Multi-task Optimization in Massively Multilingual Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Orhan Firat, Yuan Cao, Yulia Tsvetkov, Zirui Wang","submitted_at":"2020-10-12T17:26:34Z","abstract_excerpt":"Massively multilingual models subsuming tens or even hundreds of languages pose great challenges to multi-task optimization. While it is a common practice to apply a language-agnostic procedure optimizing a joint multilingual task objective, how to properly characterize and take advantage of its underlying problem structure for improving optimization efficiency remains under-explored. In this paper, we attempt to peek into the black-box of multilingual optimization through the lens of loss function geometry. We find that gradient similarity measured along the optimization trajectory is an impo"},"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":"2010.05874","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-10-12T17:26:34Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"5fccdaee9b183a6de105780ce746fdbb63ed16ca78e1b1c39c5ebb63b85b6b9d","abstract_canon_sha256":"7c54ba115050b2b2cdad4e0ac2ba2694b3c01b506bf44c10fb6e6a299d8bb1f6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T01:42:11.338754Z","signature_b64":"Uf6hHfH2ZdOoF3qJi5A4RLa/k8INkxfsDW/rJVjVNnayiw5gRtvEwTVWIhLVAMUzmtJlJBQYHn/XABUNGLA9DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a0d959ad9810b2d36f76f420cf67b3e722668cf65005c5232bf74f88ade52376","last_reissued_at":"2026-07-05T01:42:11.338281Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T01:42:11.338281Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Gradient Vaccine: Investigating and Improving Multi-task Optimization in Massively Multilingual Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Orhan Firat, Yuan Cao, Yulia Tsvetkov, Zirui Wang","submitted_at":"2020-10-12T17:26:34Z","abstract_excerpt":"Massively multilingual models subsuming tens or even hundreds of languages pose great challenges to multi-task optimization. While it is a common practice to apply a language-agnostic procedure optimizing a joint multilingual task objective, how to properly characterize and take advantage of its underlying problem structure for improving optimization efficiency remains under-explored. In this paper, we attempt to peek into the black-box of multilingual optimization through the lens of loss function geometry. We find that gradient similarity measured along the optimization trajectory is an impo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2010.05874","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/2010.05874/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":"2010.05874","created_at":"2026-07-05T01:42:11.338343+00:00"},{"alias_kind":"arxiv_version","alias_value":"2010.05874v1","created_at":"2026-07-05T01:42:11.338343+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2010.05874","created_at":"2026-07-05T01:42:11.338343+00:00"},{"alias_kind":"pith_short_12","alias_value":"UDMVTLMYCCZN","created_at":"2026-07-05T01:42:11.338343+00:00"},{"alias_kind":"pith_short_16","alias_value":"UDMVTLMYCCZNG33W","created_at":"2026-07-05T01:42:11.338343+00:00"},{"alias_kind":"pith_short_8","alias_value":"UDMVTLMY","created_at":"2026-07-05T01:42:11.338343+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.00078","citing_title":"Exploring Line Bundle Standard Models with Transformers","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17486","citing_title":"DyGRO-VLA: Cross-Task Scaling of Vision-Language-Action Models via Dynamic Grouped Residual Optimization","ref_index":115,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UDMVTLMYCCZNG33W6QQM6Z5T44","json":"https://pith.science/pith/UDMVTLMYCCZNG33W6QQM6Z5T44.json","graph_json":"https://pith.science/api/pith-number/UDMVTLMYCCZNG33W6QQM6Z5T44/graph.json","events_json":"https://pith.science/api/pith-number/UDMVTLMYCCZNG33W6QQM6Z5T44/events.json","paper":"https://pith.science/paper/UDMVTLMY"},"agent_actions":{"view_html":"https://pith.science/pith/UDMVTLMYCCZNG33W6QQM6Z5T44","download_json":"https://pith.science/pith/UDMVTLMYCCZNG33W6QQM6Z5T44.json","view_paper":"https://pith.science/paper/UDMVTLMY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2010.05874&json=true","fetch_graph":"https://pith.science/api/pith-number/UDMVTLMYCCZNG33W6QQM6Z5T44/graph.json","fetch_events":"https://pith.science/api/pith-number/UDMVTLMYCCZNG33W6QQM6Z5T44/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UDMVTLMYCCZNG33W6QQM6Z5T44/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UDMVTLMYCCZNG33W6QQM6Z5T44/action/storage_attestation","attest_author":"https://pith.science/pith/UDMVTLMYCCZNG33W6QQM6Z5T44/action/author_attestation","sign_citation":"https://pith.science/pith/UDMVTLMYCCZNG33W6QQM6Z5T44/action/citation_signature","submit_replication":"https://pith.science/pith/UDMVTLMYCCZNG33W6QQM6Z5T44/action/replication_record"}},"created_at":"2026-07-05T01:42:11.338343+00:00","updated_at":"2026-07-05T01:42:11.338343+00:00"}