{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:AXJ3PHEMXNOADFE6BFSL5K3OFC","short_pith_number":"pith:AXJ3PHEM","schema_version":"1.0","canonical_sha256":"05d3b79c8cbb5c01949e0964beab6e28abb2386bd3b966f9379d4d77c61650b6","source":{"kind":"arxiv","id":"2606.04401","version":1},"attestation_state":"computed","paper":{"title":"TANDEM: Bi-Level Data Mixture Optimization with Twin Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Deping Xiang, Guoqiang Gong, Haoran Li, Jiaxing Wang, Jin Xu, Ju Fan, Ke Zhang, Mingyang Yi, Pengzhang Liu, Qixiang Jiang, Zhen Chen, Zicheng Zhang","submitted_at":"2026-06-03T03:28:46Z","abstract_excerpt":"The capabilities of large language models (LLMs) significantly depend on training data drawn from various domains. Optimizing domain-specific mixture ratios can be modeled as a bi-level optimization problem, which we simplify into a single-level penalized form and solve with twin networks: a proxy model trained on primary data and a dynamically updated reference model trained with additional data. Our proposed method, Twin Networks for bi-level DatA mixturE optiMization (TANDEM), measures the data efficacy through the difference between the twin models and up-weights domains that benefit more "},"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":"2606.04401","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-06-03T03:28:46Z","cross_cats_sorted":[],"title_canon_sha256":"cfc6bc72f96f87add87210c49e44a0e8114b4e8cc69eaaab4148bc6a04de71c2","abstract_canon_sha256":"8d4444eec9abf9fd941ff5e3a431e6d6de3761827ae9a1d50648d7d079154e26"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-04T01:09:06.727020Z","signature_b64":"f8bq+D2xKn9/0KL1ZGMiX1hEj8exNpbK5VJMe6ny6dxtj/f0PM0IFzDT14HUczKHjl0bjhiB+ezxy9RPZWW8Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"05d3b79c8cbb5c01949e0964beab6e28abb2386bd3b966f9379d4d77c61650b6","last_reissued_at":"2026-06-04T01:09:06.726458Z","signature_status":"signed_v1","first_computed_at":"2026-06-04T01:09:06.726458Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TANDEM: Bi-Level Data Mixture Optimization with Twin Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Deping Xiang, Guoqiang Gong, Haoran Li, Jiaxing Wang, Jin Xu, Ju Fan, Ke Zhang, Mingyang Yi, Pengzhang Liu, Qixiang Jiang, Zhen Chen, Zicheng Zhang","submitted_at":"2026-06-03T03:28:46Z","abstract_excerpt":"The capabilities of large language models (LLMs) significantly depend on training data drawn from various domains. Optimizing domain-specific mixture ratios can be modeled as a bi-level optimization problem, which we simplify into a single-level penalized form and solve with twin networks: a proxy model trained on primary data and a dynamically updated reference model trained with additional data. Our proposed method, Twin Networks for bi-level DatA mixturE optiMization (TANDEM), measures the data efficacy through the difference between the twin models and up-weights domains that benefit more "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.04401","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/2606.04401/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":"2606.04401","created_at":"2026-06-04T01:09:06.726551+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.04401v1","created_at":"2026-06-04T01:09:06.726551+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.04401","created_at":"2026-06-04T01:09:06.726551+00:00"},{"alias_kind":"pith_short_12","alias_value":"AXJ3PHEMXNOA","created_at":"2026-06-04T01:09:06.726551+00:00"},{"alias_kind":"pith_short_16","alias_value":"AXJ3PHEMXNOADFE6","created_at":"2026-06-04T01:09:06.726551+00:00"},{"alias_kind":"pith_short_8","alias_value":"AXJ3PHEM","created_at":"2026-06-04T01:09:06.726551+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AXJ3PHEMXNOADFE6BFSL5K3OFC","json":"https://pith.science/pith/AXJ3PHEMXNOADFE6BFSL5K3OFC.json","graph_json":"https://pith.science/api/pith-number/AXJ3PHEMXNOADFE6BFSL5K3OFC/graph.json","events_json":"https://pith.science/api/pith-number/AXJ3PHEMXNOADFE6BFSL5K3OFC/events.json","paper":"https://pith.science/paper/AXJ3PHEM"},"agent_actions":{"view_html":"https://pith.science/pith/AXJ3PHEMXNOADFE6BFSL5K3OFC","download_json":"https://pith.science/pith/AXJ3PHEMXNOADFE6BFSL5K3OFC.json","view_paper":"https://pith.science/paper/AXJ3PHEM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.04401&json=true","fetch_graph":"https://pith.science/api/pith-number/AXJ3PHEMXNOADFE6BFSL5K3OFC/graph.json","fetch_events":"https://pith.science/api/pith-number/AXJ3PHEMXNOADFE6BFSL5K3OFC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AXJ3PHEMXNOADFE6BFSL5K3OFC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AXJ3PHEMXNOADFE6BFSL5K3OFC/action/storage_attestation","attest_author":"https://pith.science/pith/AXJ3PHEMXNOADFE6BFSL5K3OFC/action/author_attestation","sign_citation":"https://pith.science/pith/AXJ3PHEMXNOADFE6BFSL5K3OFC/action/citation_signature","submit_replication":"https://pith.science/pith/AXJ3PHEMXNOADFE6BFSL5K3OFC/action/replication_record"}},"created_at":"2026-06-04T01:09:06.726551+00:00","updated_at":"2026-06-04T01:09:06.726551+00:00"}