{"id":"https://openalex.org/W7220920765","doi":"https://doi.org/10.48550/arxiv.2610.07212","title":"Reward-Driven Learning under Prompt-Level Differential Privacy","display_name":"Reward-Driven Learning under Prompt-Level Differential Privacy","publication_year":2026,"publication_date":"2026-10-05","ids":{"openalex":"https://openalex.org/W7220920765","doi":"https://doi.org/10.48550/arxiv.2610.07212"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.07212","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.07212","pdf_url":"https://arxiv.org/pdf/2610.07212","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2610.07212","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5157982809","display_name":"Jiachen Zhao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I107639228","display_name":"University of Notre Dame","ror":"https://ror.org/00mkhxb43","country_code":"US","type":"education","lineage":["https://openalex.org/I107639228"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhao, Jiachen","raw_affiliation_strings":["University of Notre Dame"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Notre Dame","institution_ids":["https://openalex.org/I107639228"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119545355","display_name":"Antonia Januszewicz","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I107639228","display_name":"University of Notre Dame","ror":"https://ror.org/00mkhxb43","country_code":"US","type":"education","lineage":["https://openalex.org/I107639228"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Januszewicz, Antonia","raw_affiliation_strings":["University of Notre Dame"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Notre Dame","institution_ids":["https://openalex.org/I107639228"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5102709988","display_name":"Taeho Jung","orcid":"https://orcid.org/0000-0001-5873-0629","observed_orcids":["https://orcid.org/0000-0001-5873-0629"]},"institutions":[{"id":"https://openalex.org/I107639228","display_name":"University of Notre Dame","ror":"https://ror.org/00mkhxb43","country_code":"US","type":"education","lineage":["https://openalex.org/I107639228"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jung, Taeho","raw_affiliation_strings":["University of Notre Dame"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Notre Dame","institution_ids":["https://openalex.org/I107639228"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I107639228"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.9990000128746033,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.9990000128746033,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":9.999999747378752e-05,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10237","display_name":"Cryptography and Data Security","score":0.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/reinforcement-learning-with-verifiable-rewards","display_name":"reinforcement learning with verifiable rewards","score":0.8169999718666077},{"id":"https://openalex.org/keywords/gradient-aggregation","display_name":"gradient aggregation","score":0.3889999985694885},{"id":"https://openalex.org/keywords/gradient-clipping","display_name":"gradient clipping","score":0.6069999933242798},{"id":"https://openalex.org/keywords/language-model-fine-tuning","display_name":"language model fine-tuning","score":0.6830000281333923},{"id":"https://openalex.org/keywords/lora","display_name":"LoRa","score":0.5550000071525574},{"id":"https://openalex.org/keywords/math","display_name":"MATH","score":0.527999997138977},{"id":"https://openalex.org/keywords/gsm8k","display_name":"GSM8K","score":0.9919999837875366},{"id":"https://openalex.org/keywords/commonsenseqa","display_name":"CommonsenseQA","score":0.8830000162124634}],"concepts":[{"id":"https://openalex.org/C23130292","wikidata":"https://www.wikidata.org/wiki/Q5275358","display_name":"Differential privacy","level":2,"score":0.8956999778747559},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.657800018787384},{"id":"https://openalex.org/C50817715","wikidata":"https://www.wikidata.org/wiki/Q79895177","display_name":"Regret","level":2,"score":0.49880000948905945},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.49549999833106995},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.48500001430511475},{"id":"https://openalex.org/C2780615836","wikidata":"https://www.wikidata.org/wiki/Q2471869","display_name":"USable","level":2,"score":0.45170000195503235},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4366999864578247},{"id":"https://openalex.org/C2779843651","wikidata":"https://www.wikidata.org/wiki/Q7390335","display_name":"SIGNAL (programming language)","level":2,"score":0.4271000027656555},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.412200003862381},{"id":"https://openalex.org/C93226319","wikidata":"https://www.wikidata.org/wiki/Q193137","display_name":"Differential (mechanical device)","level":2,"score":0.383899986743927},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.38260000944137573},{"id":"https://openalex.org/C2776848632","wikidata":"https://www.wikidata.org/wiki/Q853463","display_name":"Clipping (morphology)","level":2,"score":0.367900013923645},{"id":"https://openalex.org/C85847156","wikidata":"https://www.wikidata.org/wiki/Q59015987","display_name":"Verifiable secret sharing","level":3,"score":0.35910001397132874},{"id":"https://openalex.org/C99221444","wikidata":"https://www.wikidata.org/wiki/Q1532069","display_name":"Private information retrieval","level":2,"score":0.3246999979019165},{"id":"https://openalex.org/C2780385302","wikidata":"https://www.wikidata.org/wiki/Q367158","display_name":"Protocol (science)","level":3,"score":0.2759999930858612}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.07212","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.07212","pdf_url":"https://arxiv.org/pdf/2610.07212","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.07212","doi":"https://doi.org/10.48550/arxiv.2610.07212","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.07212","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2610.07212","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.07212","pdf_url":"https://arxiv.org/pdf/2610.07212","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7220920765.pdf","grobid_xml":"https://content.openalex.org/works/W7220920765.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7140311031","https://openalex.org/W6891823205","https://openalex.org/W3035076532","https://openalex.org/W4415144319","https://openalex.org/W7223214679","https://openalex.org/W3006783387","https://openalex.org/W3085547302","https://openalex.org/W4378498814","https://openalex.org/W2294521904","https://openalex.org/W2784621220"],"abstract_inverted_index":{"Reinforcement":[0],"learning":[1,249],"with":[2,41,115],"verifiable":[3],"rewards":[4],"(RLVR)":[5],"trains":[6],"a":[7,118,134,171,247],"language":[8],"model":[9,20,192],"on":[10,87,125,159,163,177,197,214,218],"problems":[11,24],"that":[12,136],"may":[13],"themselves":[14],"be":[15,38],"confidential,":[16],"and":[17,76,123,129,142,161,179,199,217],"the":[18,34,44,52,60,69,78,84,89,96,104,126,131,139,154,168,187,190,209,220],"trained":[19],"can":[21],"reveal":[22],"which":[23],"it":[25],"saw.":[26],"We":[27,112],"study":[28],"RLVR":[29,110],"under":[30,251],"prompt-level":[31,252],"differential":[32,106],"privacy:":[33],"released":[35],"weights":[36],"must":[37],"(\u03b5,\u03b4)-differentially":[39],"private":[40,144,191],"respect":[42],"to":[43,56,99,203,237],"presence":[45],"of":[46,54,91,121,208,211,222],"any":[47],"one":[48,57],"training":[49],"problem.":[50],"Taking":[51],"group":[53],"responses":[55,92],"prompt":[58,94],"as":[59],"privacy":[61,79,107],"record,":[62],"our":[63,100],"method":[64],"aggregates":[65],"their":[66],"gradients,":[67],"clips":[68],"prompt's":[70],"contribution":[71],"once,":[72],"adds":[73],"Gaussian":[74],"noise,":[75],"composes":[77],"loss":[80],"across":[81],"updates,":[82],"so":[83],"budget":[85,120,189,226],"depends":[86],"neither":[88],"number":[90],"per":[93],"nor":[95],"clipping":[97],"norm;":[98],"knowledge":[101],"this":[102],"is":[103,180],"first":[105],"guarantee":[108],"for":[109],"training.":[111],"train":[113],"Qwen2.5-1.5B-Instruct":[114],"LoRA":[116],"at":[117,130,174,228],"per-run":[119],"\u03b5=8":[122],"compare,":[124],"same":[127,132,188],"prompts":[128],"budget,":[133],"control":[135,155],"removes":[137],"only":[138],"reward":[140,149,233],"signal":[141,150,250],"two":[143],"supervised":[145,195],"fine-tuning":[146],"recipes.":[147],"The":[148,232],"improves":[151],"accuracy":[152],"over":[153],"by":[156,183,201],"2.65":[157],"points":[158,176],"MATH":[160,198,219],"3.24":[162],"GSM8K,":[164],"in":[165],"every":[166],"seed;":[167],"improvement":[169],"survives":[170],"format-robust":[172],"scorer,":[173],"1.3":[175],"MATH,":[178],"not":[181],"explained":[182],"response":[184],"length.":[185],"At":[186],"outperforms":[193],"both":[194],"recipes":[196],"GSM8K":[200],"2.3":[202],"3.8":[204],"points,":[205],"retains":[206],"85--90%":[207],"gain":[210],"non-private":[212],"GRPO":[213],"these":[215],"tasks,":[216],"noise":[221],"an":[223,239],"eightfold":[224],"tighter":[225],"costs":[227],"most":[229],"1.2":[230],"points.":[231],"effect":[234],"also":[235],"carries":[236],"CommonsenseQA,":[238],"exploratory":[240],"non-mathematical":[241],"task.":[242],"Verifier":[243],"feedback":[244],"thus":[245],"remains":[246],"usable":[248],"privacy.":[253]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-10-09T00:00:00"}