{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:QXNFSYF6HORGFNBEXLKWYJF6NF","short_pith_number":"pith:QXNFSYF6","schema_version":"1.0","canonical_sha256":"85da5960be3ba262b424bad56c24be695247ffdabe960bcf7527d91799570211","source":{"kind":"arxiv","id":"2505.04560","version":3},"attestation_state":"computed","paper":{"title":"ABKD: Pursuing a Proper Allocation of the Probability Mass in Knowledge Distillation via $\\alpha$-$\\beta$-Divergence","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Guanghui Wang, Qianqian Xu, Qingming Huang, Shi Wang, Zhiyong Yang, Zitai Wang","submitted_at":"2025-05-07T16:48:49Z","abstract_excerpt":"Knowledge Distillation (KD) transfers knowledge from a large teacher model to a smaller student model by minimizing the divergence between their output distributions, typically using forward Kullback-Leibler divergence (FKLD) or reverse KLD (RKLD). It has become an effective training paradigm due to the broader supervision information provided by the teacher distribution compared to one-hot labels. We identify that the core challenge in KD lies in balancing two mode-concentration effects: the \\textbf{\\textit{Hardness-Concentration}} effect, which refers to focusing on modes with large errors, "},"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":"2505.04560","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-07T16:48:49Z","cross_cats_sorted":[],"title_canon_sha256":"f7a438060b65d08a170e3d4f0a235466bb0245792baee2e3e1ccc24fcf9276c0","abstract_canon_sha256":"6c105ff0609cb6a792bb8d88d17fd98eb759c5e3f323fa9a72ae0b3282633006"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:14:59.273908Z","signature_b64":"q/YWt06psoDRvzyCqyLVZwkd04/XAJID0eY/IT4GF+FjLE5iMcmkBIigdZIehZ+cjv6V+OnLaPr4Beboa6vTCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"85da5960be3ba262b424bad56c24be695247ffdabe960bcf7527d91799570211","last_reissued_at":"2026-07-05T11:14:59.273426Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:14:59.273426Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ABKD: Pursuing a Proper Allocation of the Probability Mass in Knowledge Distillation via $\\alpha$-$\\beta$-Divergence","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Guanghui Wang, Qianqian Xu, Qingming Huang, Shi Wang, Zhiyong Yang, Zitai Wang","submitted_at":"2025-05-07T16:48:49Z","abstract_excerpt":"Knowledge Distillation (KD) transfers knowledge from a large teacher model to a smaller student model by minimizing the divergence between their output distributions, typically using forward Kullback-Leibler divergence (FKLD) or reverse KLD (RKLD). It has become an effective training paradigm due to the broader supervision information provided by the teacher distribution compared to one-hot labels. We identify that the core challenge in KD lies in balancing two mode-concentration effects: the \\textbf{\\textit{Hardness-Concentration}} effect, which refers to focusing on modes with large errors, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.04560","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/2505.04560/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":"2505.04560","created_at":"2026-07-05T11:14:59.273485+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.04560v3","created_at":"2026-07-05T11:14:59.273485+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.04560","created_at":"2026-07-05T11:14:59.273485+00:00"},{"alias_kind":"pith_short_12","alias_value":"QXNFSYF6HORG","created_at":"2026-07-05T11:14:59.273485+00:00"},{"alias_kind":"pith_short_16","alias_value":"QXNFSYF6HORGFNBE","created_at":"2026-07-05T11:14:59.273485+00:00"},{"alias_kind":"pith_short_8","alias_value":"QXNFSYF6","created_at":"2026-07-05T11:14:59.273485+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23124","citing_title":"PRIDE: Privileged Information-enhanced Distillation for Empathetic Dialogue Generation","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03091","citing_title":"BAHSD: Bridging the Long-tail Gap via Adaptive Distillation in Black-box Sequential Recommendation","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28869","citing_title":"Balancing Multimodal Learning through Label Space Reshaping","ref_index":34,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QXNFSYF6HORGFNBEXLKWYJF6NF","json":"https://pith.science/pith/QXNFSYF6HORGFNBEXLKWYJF6NF.json","graph_json":"https://pith.science/api/pith-number/QXNFSYF6HORGFNBEXLKWYJF6NF/graph.json","events_json":"https://pith.science/api/pith-number/QXNFSYF6HORGFNBEXLKWYJF6NF/events.json","paper":"https://pith.science/paper/QXNFSYF6"},"agent_actions":{"view_html":"https://pith.science/pith/QXNFSYF6HORGFNBEXLKWYJF6NF","download_json":"https://pith.science/pith/QXNFSYF6HORGFNBEXLKWYJF6NF.json","view_paper":"https://pith.science/paper/QXNFSYF6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.04560&json=true","fetch_graph":"https://pith.science/api/pith-number/QXNFSYF6HORGFNBEXLKWYJF6NF/graph.json","fetch_events":"https://pith.science/api/pith-number/QXNFSYF6HORGFNBEXLKWYJF6NF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QXNFSYF6HORGFNBEXLKWYJF6NF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QXNFSYF6HORGFNBEXLKWYJF6NF/action/storage_attestation","attest_author":"https://pith.science/pith/QXNFSYF6HORGFNBEXLKWYJF6NF/action/author_attestation","sign_citation":"https://pith.science/pith/QXNFSYF6HORGFNBEXLKWYJF6NF/action/citation_signature","submit_replication":"https://pith.science/pith/QXNFSYF6HORGFNBEXLKWYJF6NF/action/replication_record"}},"created_at":"2026-07-05T11:14:59.273485+00:00","updated_at":"2026-07-05T11:14:59.273485+00:00"}