{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:7VW4KQZC2UI3KNYKVN3D4HPB2Q","short_pith_number":"pith:7VW4KQZC","schema_version":"1.0","canonical_sha256":"fd6dc54322d511b5370aab763e1de1d42246e99a75704e0adce070e64d1bd022","source":{"kind":"arxiv","id":"2305.13948","version":3},"attestation_state":"computed","paper":{"title":"Decoupled Kullback-Leibler Divergence Loss","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Bei Yu, Hanwang Zhang, Jiequan Cui, Xiaojuan Qi, Zhisheng Zhong, Zhuotao Tian","submitted_at":"2023-05-23T11:17:45Z","abstract_excerpt":"In this paper, we delve deeper into the Kullback-Leibler (KL) Divergence loss and mathematically prove that it is equivalent to the Decoupled Kullback-Leibler (DKL) Divergence loss that consists of 1) a weighted Mean Square Error (wMSE) loss and 2) a Cross-Entropy loss incorporating soft labels. Thanks to the decomposed formulation of DKL loss, we have identified two areas for improvement. Firstly, we address the limitation of KL/DKL in scenarios like knowledge distillation by breaking its asymmetric optimization property. This modification ensures that the $\\mathbf{w}$MSE component is always "},"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":"2305.13948","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2023-05-23T11:17:45Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"59fe5d37926d4f1b16405df91e3c41804d1616d870eee4810d174493506ad3e5","abstract_canon_sha256":"a799b568e912849a5b1b951d17a8bc7c5859a43b47f7880c1fed5f8f0b686e9b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:26:21.330523Z","signature_b64":"z1oBVVyRy7WNdQOpKQT5Wf0lMtqiprf6pYShY6dcGjl1EgUR8SUAmA5N1ssR+dvSP8QjyAaJS5hF/m/nyiQMDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fd6dc54322d511b5370aab763e1de1d42246e99a75704e0adce070e64d1bd022","last_reissued_at":"2026-07-05T09:26:21.330003Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:26:21.330003Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Decoupled Kullback-Leibler Divergence Loss","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Bei Yu, Hanwang Zhang, Jiequan Cui, Xiaojuan Qi, Zhisheng Zhong, Zhuotao Tian","submitted_at":"2023-05-23T11:17:45Z","abstract_excerpt":"In this paper, we delve deeper into the Kullback-Leibler (KL) Divergence loss and mathematically prove that it is equivalent to the Decoupled Kullback-Leibler (DKL) Divergence loss that consists of 1) a weighted Mean Square Error (wMSE) loss and 2) a Cross-Entropy loss incorporating soft labels. Thanks to the decomposed formulation of DKL loss, we have identified two areas for improvement. Firstly, we address the limitation of KL/DKL in scenarios like knowledge distillation by breaking its asymmetric optimization property. This modification ensures that the $\\mathbf{w}$MSE component is always "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.13948","kind":"arxiv","version":3},"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/2305.13948/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":"2305.13948","created_at":"2026-07-05T09:26:21.330070+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.13948v3","created_at":"2026-07-05T09:26:21.330070+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.13948","created_at":"2026-07-05T09:26:21.330070+00:00"},{"alias_kind":"pith_short_12","alias_value":"7VW4KQZC2UI3","created_at":"2026-07-05T09:26:21.330070+00:00"},{"alias_kind":"pith_short_16","alias_value":"7VW4KQZC2UI3KNYK","created_at":"2026-07-05T09:26:21.330070+00:00"},{"alias_kind":"pith_short_8","alias_value":"7VW4KQZC","created_at":"2026-07-05T09:26:21.330070+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2512.10275","citing_title":"Sample-wise Adaptive Weighting for Transfer Consistency in Adversarial Distillation","ref_index":20,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7VW4KQZC2UI3KNYKVN3D4HPB2Q","json":"https://pith.science/pith/7VW4KQZC2UI3KNYKVN3D4HPB2Q.json","graph_json":"https://pith.science/api/pith-number/7VW4KQZC2UI3KNYKVN3D4HPB2Q/graph.json","events_json":"https://pith.science/api/pith-number/7VW4KQZC2UI3KNYKVN3D4HPB2Q/events.json","paper":"https://pith.science/paper/7VW4KQZC"},"agent_actions":{"view_html":"https://pith.science/pith/7VW4KQZC2UI3KNYKVN3D4HPB2Q","download_json":"https://pith.science/pith/7VW4KQZC2UI3KNYKVN3D4HPB2Q.json","view_paper":"https://pith.science/paper/7VW4KQZC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.13948&json=true","fetch_graph":"https://pith.science/api/pith-number/7VW4KQZC2UI3KNYKVN3D4HPB2Q/graph.json","fetch_events":"https://pith.science/api/pith-number/7VW4KQZC2UI3KNYKVN3D4HPB2Q/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7VW4KQZC2UI3KNYKVN3D4HPB2Q/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7VW4KQZC2UI3KNYKVN3D4HPB2Q/action/storage_attestation","attest_author":"https://pith.science/pith/7VW4KQZC2UI3KNYKVN3D4HPB2Q/action/author_attestation","sign_citation":"https://pith.science/pith/7VW4KQZC2UI3KNYKVN3D4HPB2Q/action/citation_signature","submit_replication":"https://pith.science/pith/7VW4KQZC2UI3KNYKVN3D4HPB2Q/action/replication_record"}},"created_at":"2026-07-05T09:26:21.330070+00:00","updated_at":"2026-07-05T09:26:21.330070+00:00"}