{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2018:C5TUS6HMELPFDBW7XSO3XS3OSO","short_pith_number":"pith:C5TUS6HM","schema_version":"1.0","canonical_sha256":"17674978ec22de5186dfbc9dbbcb6e93ad283122c955b18ad0ebfa7335951096","source":{"kind":"arxiv","id":"1812.02224","version":2},"attestation_state":"computed","paper":{"title":"Adapting Auxiliary Losses Using Gradient Similarity","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Balaji Lakshminarayanan, Mehrdad Farajtabar, Razvan Pascanu, Siddhant M. Jayakumar, Wojciech M. Czarnecki, Yunshu Du","submitted_at":"2018-12-05T21:00:44Z","abstract_excerpt":"One approach to deal with the statistical inefficiency of neural networks is to rely on auxiliary losses that help to build useful representations. However, it is not always trivial to know if an auxiliary task will be helpful for the main task and when it could start hurting. We propose to use the cosine similarity between gradients of tasks as an adaptive weight to detect when an auxiliary loss is helpful to the main loss. We show that our approach is guaranteed to converge to critical points of the main task and demonstrate the practical usefulness of the proposed algorithm in a few domains"},"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":"1812.02224","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2018-12-05T21:00:44Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"fd007adb59d6a987b9d74f17bb5fa345931aefd0f6b217e31639fcf9f995eddd","abstract_canon_sha256":"b8b9f17b63868b170481a76be2c85f73bdf2b7028499ca0cf2e005b819cf19f8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T01:54:39.568310Z","signature_b64":"9hVWvvccN9qTBoWSWzUxbw8EJayQ//VfSOul7D91aTVSjC53CYRUYoBSZcFY13KXaIDMUMt6RXCyVBrf4N8FCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"17674978ec22de5186dfbc9dbbcb6e93ad283122c955b18ad0ebfa7335951096","last_reissued_at":"2026-07-05T01:54:39.567848Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T01:54:39.567848Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Adapting Auxiliary Losses Using Gradient Similarity","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Balaji Lakshminarayanan, Mehrdad Farajtabar, Razvan Pascanu, Siddhant M. Jayakumar, Wojciech M. Czarnecki, Yunshu Du","submitted_at":"2018-12-05T21:00:44Z","abstract_excerpt":"One approach to deal with the statistical inefficiency of neural networks is to rely on auxiliary losses that help to build useful representations. However, it is not always trivial to know if an auxiliary task will be helpful for the main task and when it could start hurting. We propose to use the cosine similarity between gradients of tasks as an adaptive weight to detect when an auxiliary loss is helpful to the main loss. We show that our approach is guaranteed to converge to critical points of the main task and demonstrate the practical usefulness of the proposed algorithm in a few domains"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1812.02224","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/1812.02224/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":"1812.02224","created_at":"2026-07-05T01:54:39.567905+00:00"},{"alias_kind":"arxiv_version","alias_value":"1812.02224v2","created_at":"2026-07-05T01:54:39.567905+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1812.02224","created_at":"2026-07-05T01:54:39.567905+00:00"},{"alias_kind":"pith_short_12","alias_value":"C5TUS6HMELPF","created_at":"2026-07-05T01:54:39.567905+00:00"},{"alias_kind":"pith_short_16","alias_value":"C5TUS6HMELPFDBW7","created_at":"2026-07-05T01:54:39.567905+00:00"},{"alias_kind":"pith_short_8","alias_value":"C5TUS6HM","created_at":"2026-07-05T01:54:39.567905+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.25001","citing_title":"Mitigating Gradient Pathology in PINNs through Aligned Constraint","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30011","citing_title":"T3R: Deeper Test-Time Adaptation for Graph Neural Networks via Gradient Rotation","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2402.08267","citing_title":"Improving Image Coding for Machines through Optimizing Encoder via Auxiliary Loss","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20771","citing_title":"Cumulative Meta-Learning from Active Learning Queries for Robustness to Spurious Correlations","ref_index":133,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07756","citing_title":"When Losses Align: Gradient-Based Composite Loss Weighting for Efficient Pretraining","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16836","citing_title":"Lorentz Framework for Semantic Segmentation","ref_index":17,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/C5TUS6HMELPFDBW7XSO3XS3OSO","json":"https://pith.science/pith/C5TUS6HMELPFDBW7XSO3XS3OSO.json","graph_json":"https://pith.science/api/pith-number/C5TUS6HMELPFDBW7XSO3XS3OSO/graph.json","events_json":"https://pith.science/api/pith-number/C5TUS6HMELPFDBW7XSO3XS3OSO/events.json","paper":"https://pith.science/paper/C5TUS6HM"},"agent_actions":{"view_html":"https://pith.science/pith/C5TUS6HMELPFDBW7XSO3XS3OSO","download_json":"https://pith.science/pith/C5TUS6HMELPFDBW7XSO3XS3OSO.json","view_paper":"https://pith.science/paper/C5TUS6HM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1812.02224&json=true","fetch_graph":"https://pith.science/api/pith-number/C5TUS6HMELPFDBW7XSO3XS3OSO/graph.json","fetch_events":"https://pith.science/api/pith-number/C5TUS6HMELPFDBW7XSO3XS3OSO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/C5TUS6HMELPFDBW7XSO3XS3OSO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/C5TUS6HMELPFDBW7XSO3XS3OSO/action/storage_attestation","attest_author":"https://pith.science/pith/C5TUS6HMELPFDBW7XSO3XS3OSO/action/author_attestation","sign_citation":"https://pith.science/pith/C5TUS6HMELPFDBW7XSO3XS3OSO/action/citation_signature","submit_replication":"https://pith.science/pith/C5TUS6HMELPFDBW7XSO3XS3OSO/action/replication_record"}},"created_at":"2026-07-05T01:54:39.567905+00:00","updated_at":"2026-07-05T01:54:39.567905+00:00"}