{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:S5UAZ64FM5T5LD3SZU6EG2VMZX","short_pith_number":"pith:S5UAZ64F","canonical_record":{"source":{"id":"2605.13149","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T08:15:48Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"2752767d01801f55bda50326841fb7219eaa66a04fc96bf2521df3d9ef580de2","abstract_canon_sha256":"744050bae97cfb73e7ba0071880c23b80262a0161883a193b4ac1cb782cf66b0"},"schema_version":"1.0"},"canonical_sha256":"97680cfb856767d58f72cd3c436aaccdc6b06558d9980ce9bd96609ac6e9f8ff","source":{"kind":"arxiv","id":"2605.13149","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13149","created_at":"2026-05-18T03:08:57Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13149v1","created_at":"2026-05-18T03:08:57Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13149","created_at":"2026-05-18T03:08:57Z"},{"alias_kind":"pith_short_12","alias_value":"S5UAZ64FM5T5","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"S5UAZ64FM5T5LD3S","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"S5UAZ64F","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:S5UAZ64FM5T5LD3SZU6EG2VMZX","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13149","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T08:15:48Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"2752767d01801f55bda50326841fb7219eaa66a04fc96bf2521df3d9ef580de2","abstract_canon_sha256":"744050bae97cfb73e7ba0071880c23b80262a0161883a193b4ac1cb782cf66b0"},"schema_version":"1.0"},"canonical_sha256":"97680cfb856767d58f72cd3c436aaccdc6b06558d9980ce9bd96609ac6e9f8ff","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:08:57.208690Z","signature_b64":"y3zk0Vxh9q2cY0yDehe0Qp1+0676XlkANFHhhknczqJAQnyt9jV6H157Ng9PCY8S8FgaB73/cIn5OnlUFDN9Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"97680cfb856767d58f72cd3c436aaccdc6b06558d9980ce9bd96609ac6e9f8ff","last_reissued_at":"2026-05-18T03:08:57.208072Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:08:57.208072Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13149","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:08:57Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"CLrpBgMK8U22T2Das2CV9bxAGmyQFtaOfWKdVIHQuhvlyvjBjsDKFm/SJf+5dNEs+41jT+1uxol3iHWkSuQaCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-30T02:24:37.965297Z"},"content_sha256":"7fabd97372ccac6d16e56dd9d517e59bdff955b479d84a78655a9abb06d00b47","schema_version":"1.0","event_id":"sha256:7fabd97372ccac6d16e56dd9d517e59bdff955b479d84a78655a9abb06d00b47"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:S5UAZ64FM5T5LD3SZU6EG2VMZX","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"AcquisitionSynthesis: Targeted Data Generation using Acquisition Functions","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Acquisition functions as rewards train models to generate synthetic data that improves student performance by 2-7 percent","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Dilek Hakkani-T\\\"ur, Emre Can Acikgoz, Ishika Agarwal, Jiaqi Ma, Mahdi Namazifar, Pradeep Natarajan, Sofia Stoica","submitted_at":"2026-05-13T08:15:48Z","abstract_excerpt":"Data quality remains a critical bottleneck in developing capable, competitive models. Researchers have explored many ways to generate top quality samples. Some works rely on rejection sampling: generating lots of synthetic samples and filtering out low-quality samples. Other works rely on larger or closed-source models to extract model weaknesses, necessary skills, or a curriculum off of which to base data generation. These works have one common limitation: there is no quantitative approach to measure the impact of the generated samples on the downstream learner. Active learning literature pro"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"student models trained with AcquisitionSynthesis data achieve good performance on in-distribution tasks (2-7% gain) and is more robust to catastrophic forgetting","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That acquisition functions provide a reliable quantitative signal of a sample's impact on the downstream learner that can be directly used as a reward without further validation or adaptation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"AcquisitionSynthesis uses acquisition functions as rewards to train generators that produce higher-quality synthetic data, delivering 2-7% gains on math, medical QA, and coding tasks with improved robustness to forgetting.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Acquisition functions as rewards train models to generate synthetic data that improves student performance by 2-7 percent","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"eb8b5488ce498a0039b919033e9360a8a5814f060de4293d4ddb2e1b9a2292fd"},"source":{"id":"2605.13149","kind":"arxiv","version":1},"verdict":{"id":"7f88e0d1-18fe-4abd-b715-d1f59e93b36c","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:26:22.803140Z","strongest_claim":"student models trained with AcquisitionSynthesis data achieve good performance on in-distribution tasks (2-7% gain) and is more robust to catastrophic forgetting","one_line_summary":"AcquisitionSynthesis uses acquisition functions as rewards to train generators that produce higher-quality synthetic data, delivering 2-7% gains on math, medical QA, and coding tasks with improved robustness to forgetting.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That acquisition functions provide a reliable quantitative signal of a sample's impact on the downstream learner that can be directly used as a reward without further validation or adaptation.","pith_extraction_headline":"Acquisition functions as rewards train models to generate synthetic data that improves student performance by 2-7 percent"},"references":{"count":62,"sample":[{"doi":"","year":2020,"title":"Understanding Black-box Predictions via Influence Functions , author=. 2020 , eprint=","work_id":"a9dfb044-c0e1-4511-8540-7c46071e493b","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"DataEnvGym: Data Generation Agents in Teacher Environments with Student Feedback , author=. 2025 , eprint=","work_id":"1ac7f0ba-cb70-46bc-a120-a6faa50a90c2","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Xia, Mengzhou and Malladi, Sadhika and Gururangan, Suchin and Arora, Sanjeev and Chen, Danqi , booktitle=","work_id":"c92cd0db-7c1d-400e-8801-9c1d9757c0a2","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Reflexion: Language Agents with Verbal Reinforcement Learning , author=. 2023 , eprint=","work_id":"9e726140-07d9-49d4-b4fc-1ce0d03d41ee","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"DAPO: An Open-Source LLM Reinforcement Learning System at Scale , author=. 2025 , eprint=","work_id":"23c9ed8b-0801-487e-b54b-89ecae58740e","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":62,"snapshot_sha256":"19a54c71aeb876930d3fecc0ed768972b979f8ee0892b41576e44946ec3bba30","internal_anchors":2},"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"},"verdict_id":"7f88e0d1-18fe-4abd-b715-d1f59e93b36c"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:08:57Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"r7qk9563ECZzP0/KqvLdFC7Zk8huz3M/aYBv/tfTrWTjANFOP6U1fYPivZmPeyOfBo1R6WW3fG+zT7iFqAX6Bw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-30T02:24:37.966293Z"},"content_sha256":"d14684c632ea62657943e68a2adde0ae23883be1e14c65b7fc3ce3f0f62c9b21","schema_version":"1.0","event_id":"sha256:d14684c632ea62657943e68a2adde0ae23883be1e14c65b7fc3ce3f0f62c9b21"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/S5UAZ64FM5T5LD3SZU6EG2VMZX/bundle.json","state_url":"https://pith.science/pith/S5UAZ64FM5T5LD3SZU6EG2VMZX/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/S5UAZ64FM5T5LD3SZU6EG2VMZX/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-30T02:24:37Z","links":{"resolver":"https://pith.science/pith/S5UAZ64FM5T5LD3SZU6EG2VMZX","bundle":"https://pith.science/pith/S5UAZ64FM5T5LD3SZU6EG2VMZX/bundle.json","state":"https://pith.science/pith/S5UAZ64FM5T5LD3SZU6EG2VMZX/state.json","well_known_bundle":"https://pith.science/.well-known/pith/S5UAZ64FM5T5LD3SZU6EG2VMZX/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:S5UAZ64FM5T5LD3SZU6EG2VMZX","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"744050bae97cfb73e7ba0071880c23b80262a0161883a193b4ac1cb782cf66b0","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T08:15:48Z","title_canon_sha256":"2752767d01801f55bda50326841fb7219eaa66a04fc96bf2521df3d9ef580de2"},"schema_version":"1.0","source":{"id":"2605.13149","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13149","created_at":"2026-05-18T03:08:57Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13149v1","created_at":"2026-05-18T03:08:57Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13149","created_at":"2026-05-18T03:08:57Z"},{"alias_kind":"pith_short_12","alias_value":"S5UAZ64FM5T5","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"S5UAZ64FM5T5LD3S","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"S5UAZ64F","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:d14684c632ea62657943e68a2adde0ae23883be1e14c65b7fc3ce3f0f62c9b21","target":"graph","created_at":"2026-05-18T03:08:57Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"student models trained with AcquisitionSynthesis data achieve good performance on in-distribution tasks (2-7% gain) and is more robust to catastrophic forgetting"},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That acquisition functions provide a reliable quantitative signal of a sample's impact on the downstream learner that can be directly used as a reward without further validation or adaptation."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"AcquisitionSynthesis uses acquisition functions as rewards to train generators that produce higher-quality synthetic data, delivering 2-7% gains on math, medical QA, and coding tasks with improved robustness to forgetting."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Acquisition functions as rewards train models to generate synthetic data that improves student performance by 2-7 percent"}],"snapshot_sha256":"eb8b5488ce498a0039b919033e9360a8a5814f060de4293d4ddb2e1b9a2292fd"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Data quality remains a critical bottleneck in developing capable, competitive models. Researchers have explored many ways to generate top quality samples. Some works rely on rejection sampling: generating lots of synthetic samples and filtering out low-quality samples. Other works rely on larger or closed-source models to extract model weaknesses, necessary skills, or a curriculum off of which to base data generation. These works have one common limitation: there is no quantitative approach to measure the impact of the generated samples on the downstream learner. Active learning literature pro","authors_text":"Dilek Hakkani-T\\\"ur, Emre Can Acikgoz, Ishika Agarwal, Jiaqi Ma, Mahdi Namazifar, Pradeep Natarajan, Sofia Stoica","cross_cats":["cs.AI","cs.LG"],"headline":"Acquisition functions as rewards train models to generate synthetic data that improves student performance by 2-7 percent","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T08:15:48Z","title":"AcquisitionSynthesis: Targeted Data Generation using Acquisition Functions"},"references":{"count":62,"internal_anchors":2,"resolved_work":62,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Understanding Black-box Predictions via Influence Functions , author=. 2020 , eprint=","work_id":"a9dfb044-c0e1-4511-8540-7c46071e493b","year":2020},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"DataEnvGym: Data Generation Agents in Teacher Environments with Student Feedback , author=. 2025 , eprint=","work_id":"1ac7f0ba-cb70-46bc-a120-a6faa50a90c2","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Xia, Mengzhou and Malladi, Sadhika and Gururangan, Suchin and Arora, Sanjeev and Chen, Danqi , booktitle=","work_id":"c92cd0db-7c1d-400e-8801-9c1d9757c0a2","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Reflexion: Language Agents with Verbal Reinforcement Learning , author=. 2023 , eprint=","work_id":"9e726140-07d9-49d4-b4fc-1ce0d03d41ee","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"DAPO: An Open-Source LLM Reinforcement Learning System at Scale , author=. 2025 , eprint=","work_id":"23c9ed8b-0801-487e-b54b-89ecae58740e","year":2025}],"snapshot_sha256":"19a54c71aeb876930d3fecc0ed768972b979f8ee0892b41576e44946ec3bba30"},"source":{"id":"2605.13149","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T19:26:22.803140Z","id":"7f88e0d1-18fe-4abd-b715-d1f59e93b36c","model_set":{"reader":"grok-4.3"},"one_line_summary":"AcquisitionSynthesis uses acquisition functions as rewards to train generators that produce higher-quality synthetic data, delivering 2-7% gains on math, medical QA, and coding tasks with improved robustness to forgetting.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Acquisition functions as rewards train models to generate synthetic data that improves student performance by 2-7 percent","strongest_claim":"student models trained with AcquisitionSynthesis data achieve good performance on in-distribution tasks (2-7% gain) and is more robust to catastrophic forgetting","weakest_assumption":"That acquisition functions provide a reliable quantitative signal of a sample's impact on the downstream learner that can be directly used as a reward without further validation or adaptation."}},"verdict_id":"7f88e0d1-18fe-4abd-b715-d1f59e93b36c"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:7fabd97372ccac6d16e56dd9d517e59bdff955b479d84a78655a9abb06d00b47","target":"record","created_at":"2026-05-18T03:08:57Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"744050bae97cfb73e7ba0071880c23b80262a0161883a193b4ac1cb782cf66b0","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T08:15:48Z","title_canon_sha256":"2752767d01801f55bda50326841fb7219eaa66a04fc96bf2521df3d9ef580de2"},"schema_version":"1.0","source":{"id":"2605.13149","kind":"arxiv","version":1}},"canonical_sha256":"97680cfb856767d58f72cd3c436aaccdc6b06558d9980ce9bd96609ac6e9f8ff","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"97680cfb856767d58f72cd3c436aaccdc6b06558d9980ce9bd96609ac6e9f8ff","first_computed_at":"2026-05-18T03:08:57.208072Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:08:57.208072Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"y3zk0Vxh9q2cY0yDehe0Qp1+0676XlkANFHhhknczqJAQnyt9jV6H157Ng9PCY8S8FgaB73/cIn5OnlUFDN9Dg==","signature_status":"signed_v1","signed_at":"2026-05-18T03:08:57.208690Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13149","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:7fabd97372ccac6d16e56dd9d517e59bdff955b479d84a78655a9abb06d00b47","sha256:d14684c632ea62657943e68a2adde0ae23883be1e14c65b7fc3ce3f0f62c9b21"],"state_sha256":"e1fb4b574dfcc597d9b6523de69c9c49363a79c44116ddb6db7758e14354ca95"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"p46lHdTn3YrB5CZt8s2E8lEIqEFsAxdY5JlqojHalOpErbUXX8N8n3T8es5sBnCUCx9/cMQ1ZKQc1eE6V5jEDA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-30T02:24:37.970575Z","bundle_sha256":"981587d1f26b8b20b88baa9934f27ae11e0b961e8707626993aaa5f0763fdc1b"}}