{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:2AYEGCXOY45PVPQHIXXNAHSLAQ","short_pith_number":"pith:2AYEGCXO","schema_version":"1.0","canonical_sha256":"d030430aeec73afabe0745eed01e4b04355a463dd3db1aefa6e2b828c01f03cf","source":{"kind":"arxiv","id":"2108.11346","version":1},"attestation_state":"computed","paper":{"title":"Auxiliary Task Update Decomposition: The Good, The Bad and The Neutral","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"David Grangier, Lucio M. Dery, Yann Dauphin","submitted_at":"2021-08-25T17:09:48Z","abstract_excerpt":"While deep learning has been very beneficial in data-rich settings, tasks with smaller training set often resort to pre-training or multitask learning to leverage data from other tasks. In this case, careful consideration is needed to select tasks and model parameterizations such that updates from the auxiliary tasks actually help the primary task. We seek to alleviate this burden by formulating a model-agnostic framework that performs fine-grained manipulation of the auxiliary task gradients. We propose to decompose auxiliary updates into directions which help, damage or leave the primary tas"},"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":"2108.11346","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-08-25T17:09:48Z","cross_cats_sorted":[],"title_canon_sha256":"988f8c3f43926bf49d00edeb7982adc435d2054b76e21ac7ea3f2a7c79a71c02","abstract_canon_sha256":"47edf921220aa0839ae3b0b1adccd27c391b40d7ef60771ace60f598e0774884"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:08:52.684541Z","signature_b64":"ee8/afokFLkqSw0GSlfOqfNG9b3wBwcF477tzmU2R3YXTAtP9Xjt1S+W7eOnseuO62o4eiAeidDz7P7Xr8ZJCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d030430aeec73afabe0745eed01e4b04355a463dd3db1aefa6e2b828c01f03cf","last_reissued_at":"2026-07-05T03:08:52.684188Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:08:52.684188Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Auxiliary Task Update Decomposition: The Good, The Bad and The Neutral","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"David Grangier, Lucio M. Dery, Yann Dauphin","submitted_at":"2021-08-25T17:09:48Z","abstract_excerpt":"While deep learning has been very beneficial in data-rich settings, tasks with smaller training set often resort to pre-training or multitask learning to leverage data from other tasks. In this case, careful consideration is needed to select tasks and model parameterizations such that updates from the auxiliary tasks actually help the primary task. We seek to alleviate this burden by formulating a model-agnostic framework that performs fine-grained manipulation of the auxiliary task gradients. We propose to decompose auxiliary updates into directions which help, damage or leave the primary tas"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2108.11346","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/2108.11346/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":"2108.11346","created_at":"2026-07-05T03:08:52.684247+00:00"},{"alias_kind":"arxiv_version","alias_value":"2108.11346v1","created_at":"2026-07-05T03:08:52.684247+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2108.11346","created_at":"2026-07-05T03:08:52.684247+00:00"},{"alias_kind":"pith_short_12","alias_value":"2AYEGCXOY45P","created_at":"2026-07-05T03:08:52.684247+00:00"},{"alias_kind":"pith_short_16","alias_value":"2AYEGCXOY45PVPQH","created_at":"2026-07-05T03:08:52.684247+00:00"},{"alias_kind":"pith_short_8","alias_value":"2AYEGCXO","created_at":"2026-07-05T03:08:52.684247+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2502.02067","citing_title":"AdaptBot: Combining LLM with Knowledge Graphs and Human Input for Generic-to-Specific Task Decomposition and Knowledge Refinement","ref_index":14,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2AYEGCXOY45PVPQHIXXNAHSLAQ","json":"https://pith.science/pith/2AYEGCXOY45PVPQHIXXNAHSLAQ.json","graph_json":"https://pith.science/api/pith-number/2AYEGCXOY45PVPQHIXXNAHSLAQ/graph.json","events_json":"https://pith.science/api/pith-number/2AYEGCXOY45PVPQHIXXNAHSLAQ/events.json","paper":"https://pith.science/paper/2AYEGCXO"},"agent_actions":{"view_html":"https://pith.science/pith/2AYEGCXOY45PVPQHIXXNAHSLAQ","download_json":"https://pith.science/pith/2AYEGCXOY45PVPQHIXXNAHSLAQ.json","view_paper":"https://pith.science/paper/2AYEGCXO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2108.11346&json=true","fetch_graph":"https://pith.science/api/pith-number/2AYEGCXOY45PVPQHIXXNAHSLAQ/graph.json","fetch_events":"https://pith.science/api/pith-number/2AYEGCXOY45PVPQHIXXNAHSLAQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2AYEGCXOY45PVPQHIXXNAHSLAQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2AYEGCXOY45PVPQHIXXNAHSLAQ/action/storage_attestation","attest_author":"https://pith.science/pith/2AYEGCXOY45PVPQHIXXNAHSLAQ/action/author_attestation","sign_citation":"https://pith.science/pith/2AYEGCXOY45PVPQHIXXNAHSLAQ/action/citation_signature","submit_replication":"https://pith.science/pith/2AYEGCXOY45PVPQHIXXNAHSLAQ/action/replication_record"}},"created_at":"2026-07-05T03:08:52.684247+00:00","updated_at":"2026-07-05T03:08:52.684247+00:00"}