{"id":"https://openalex.org/W7214995033","doi":"https://doi.org/10.48550/arxiv.2609.36264","title":"OTROPE: Optimal Transport-based Robust Off-policy Evaluation for Large Language Models","display_name":"OTROPE: Optimal Transport-based Robust Off-policy Evaluation for Large Language Models","publication_year":2026,"publication_date":"2026-09-28","ids":{"openalex":"https://openalex.org/W7214995033","doi":"https://doi.org/10.48550/arxiv.2609.36264"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.36264","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.36264","pdf_url":"https://arxiv.org/pdf/2609.36264","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.36264","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5153620238","display_name":"Liner Xiang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I204250578","display_name":"University of California, Irvine","ror":"https://ror.org/04gyf1771","country_code":"US","type":"education","lineage":["https://openalex.org/I204250578"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Xiang, Liner","raw_affiliation_strings":["Department of Statistics , University of California , Irvine"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Statistics , University of California , Irvine","institution_ids":["https://openalex.org/I204250578"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153656416","display_name":"Wenbo Zhang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I204250578","display_name":"University of California, Irvine","ror":"https://ror.org/04gyf1771","country_code":"US","type":"education","lineage":["https://openalex.org/I204250578"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhang, Wenbo","raw_affiliation_strings":["Department of Statistics , University of California , Irvine"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Statistics , University of California , Irvine","institution_ids":["https://openalex.org/I204250578"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5072907251","display_name":"Hengrui Cai","orcid":"https://orcid.org/0000-0002-5679-8862","observed_orcids":["https://orcid.org/0000-0002-5679-8862"]},"institutions":[{"id":"https://openalex.org/I204250578","display_name":"University of California, Irvine","ror":"https://ror.org/04gyf1771","country_code":"US","type":"education","lineage":["https://openalex.org/I204250578"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Cai, Hengrui","raw_affiliation_strings":["Department of Statistics , University of California , Irvine"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Statistics , University of California , Irvine","institution_ids":["https://openalex.org/I204250578"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I204250578"],"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.43380001187324524,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.43380001187324524,"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/T12101","display_name":"Advanced Bandit Algorithms Research","score":0.14110000431537628,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10845","display_name":"Advanced Causal Inference Techniques","score":0.036400001496076584,"subfield":{"id":"https://openalex.org/subfields/2613","display_name":"Statistics and Probability"},"field":{"id":"https://openalex.org/fields/26","display_name":"Mathematics"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/off-policy-evaluation","display_name":"off-policy evaluation","score":0.8970000147819519},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9950000047683716},{"id":"https://openalex.org/keywords/optimal-transport","display_name":"optimal transport","score":0.9769999980926514},{"id":"https://openalex.org/keywords/distribution-shift","display_name":"distribution shift","score":0.7310000061988831},{"id":"https://openalex.org/keywords/doubly-robust-estimation","display_name":"doubly robust estimation","score":0.6240000128746033},{"id":"https://openalex.org/keywords/llm-evaluation","display_name":"LLM evaluation","score":0.5680000185966492}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.697700023651123},{"id":"https://openalex.org/C2776436953","wikidata":"https://www.wikidata.org/wiki/Q5163215","display_name":"Consistency (knowledge bases)","level":2,"score":0.6029999852180481},{"id":"https://openalex.org/C2780148112","wikidata":"https://www.wikidata.org/wiki/Q1432581","display_name":"Proxy (statistics)","level":2,"score":0.5110999941825867},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4787999987602234},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4487999975681305},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.4406000077724457},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.41190001368522644},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.3813000023365021},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.35260000824928284},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.33980000019073486},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.3156999945640564},{"id":"https://openalex.org/C67226441","wikidata":"https://www.wikidata.org/wiki/Q1665389","display_name":"Robust statistics","level":3,"score":0.30550000071525574},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.3018999993801117},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.3000999987125397},{"id":"https://openalex.org/C2776135515","wikidata":"https://www.wikidata.org/wiki/Q17143721","display_name":"Regularization (linguistics)","level":2,"score":0.27379998564720154},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.2711000144481659},{"id":"https://openalex.org/C2780598303","wikidata":"https://www.wikidata.org/wiki/Q65921492","display_name":"Flexibility (engineering)","level":2,"score":0.2619999945163727},{"id":"https://openalex.org/C3019813237","wikidata":"https://www.wikidata.org/wiki/Q65089264","display_name":"Model validation","level":2,"score":0.2531000077724457}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.36264","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.36264","pdf_url":"https://arxiv.org/pdf/2609.36264","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.36264","doi":"https://doi.org/10.48550/arxiv.2609.36264","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.36264","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:2609.36264","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.36264","pdf_url":"https://arxiv.org/pdf/2609.36264","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G4454352008","display_name":"Collaborative Research: Causal Discovery and Individualized Policy Optimization for Human Text Data","funder_award_id":"2401271","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214995033.pdf","grobid_xml":"https://content.openalex.org/works/W7214995033.grobid-xml"},"referenced_works_count":13,"referenced_works":["https://openalex.org/W4386907945","https://openalex.org/W4388327685","https://openalex.org/W4389519254","https://openalex.org/W4392781612","https://openalex.org/W4399759325","https://openalex.org/W4402670856","https://openalex.org/W4403885225","https://openalex.org/W4404347031","https://openalex.org/W4405562067","https://openalex.org/W4405903187","https://openalex.org/W4416255968","https://openalex.org/W6947813222","https://openalex.org/W7147676018"],"related_works":["https://openalex.org/W7221076092","https://openalex.org/W4403884617","https://openalex.org/W7126231948","https://openalex.org/W7161163974","https://openalex.org/W7169671228","https://openalex.org/W4394838611","https://openalex.org/W4396777837","https://openalex.org/W4382173247","https://openalex.org/W4407684013","https://openalex.org/W4415939116"],"abstract_inverted_index":{"Reliable":[0],"evaluation":[1,28,80,113,156],"of":[2,167],"large":[3],"language":[4],"models":[5],"(LLMs)":[6],"is":[7,15,49,58,179],"essential":[8],"for":[9,29,66,137],"their":[10],"development":[11],"and":[12,19,60,131,134,153,173],"deployment,":[13],"yet":[14],"often":[16,64],"costly,":[17],"risky,":[18],"difficult":[20],"to":[21,41,92,171],"perform":[22],"safely":[23],"online.":[24],"We":[25,69,120],"study":[26],"off-policy":[27],"LLMs,":[30],"where":[31],"limited":[32],"human-labeled":[33,104],"data":[34],"from":[35],"a":[36,43,78,86,110],"behavior":[37,143],"model":[38],"are":[39,53,63],"used":[40],"evaluate":[42],"newer":[44],"target":[45],"LLM.":[46],"This":[47],"setting":[48],"challenging":[50],"because":[51],"labels":[52],"scarce,":[54],"behavior--target":[55],"distribution":[56,129,144],"shift":[57],"common,":[59],"response":[61],"likelihoods":[62],"unavailable":[65],"black-box":[67],"LLMs.":[68],"propose":[70],"the":[71,141,146],"Optimal":[72],"Transport-based":[73],"Robust":[74],"Off-Policy":[75],"Evaluation":[76],"(OTROPE),":[77],"likelihood-free":[79],"that":[81,159],"performs":[82],"distributional":[83],"correction":[84],"in":[85],"semantic":[87],"space":[88],"via":[89],"optimal":[90],"transport":[91],"align":[93],"labeled":[94],"behavior-policy":[95,115],"samples":[96],"with":[97,106],"unlabeled":[98],"target-policy":[99],"samples.":[100],"OTROPE":[101,138,160],"combines":[102],"corrected":[103],"residuals":[105],"proxy":[107,147],"predictors,":[108],"yielding":[109],"doubly":[111],"robust-style":[112],"without":[114],"modeling":[116],"or":[117,145],"density-ratio":[118],"estimation.":[119],"theoretically":[121],"characterize":[122],"why":[123],"baseline":[124],"evaluators":[125,170],"fail":[126],"under":[127],"LLM":[128,155,169],"shift,":[130],"establish":[132],"consistency":[133],"convergence":[135],"rates":[136],"when":[139],"either":[140],"reweighted":[142],"predictor":[148],"converges.":[149],"Experiments":[150],"on":[151],"synthetic":[152],"real":[154],"tasks":[157],"show":[158],"consistently":[161],"outperforms":[162],"baselines":[163],"while":[164],"enabling":[165],"ensembles":[166],"weaker":[168],"approach":[172],"sometimes":[174],"surpass":[175],"stronger":[176],"evaluators.":[177],"Code":[178],"available":[180],"at":[181],"https://github.com/LinerXiang/OTROPE.":[182]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-01T00:00:00"}