{"id":"https://openalex.org/W7139936135","doi":"https://doi.org/10.48550/arxiv.2603.18736","title":"CausalRM: Causal-Theoretic Reward Modeling for RLHF from Observational User Feedbacks","display_name":"CausalRM: Causal-Theoretic Reward Modeling for RLHF from Observational User Feedbacks","publication_year":2026,"publication_date":"2026-03-19","ids":{"openalex":"https://openalex.org/W7139936135","doi":"https://doi.org/10.48550/arxiv.2603.18736"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.18736","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.18736","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.18736","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5130215411","display_name":"Hao Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Hao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124278084","display_name":"Licheng Pan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pan, Licheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130243784","display_name":"Zhichao Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Zhichao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130240648","display_name":"Chunyuan Zheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng, Chunyuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130230076","display_name":"Zhixuan Chu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chu, Zhixuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5127365116","display_name":"Xiaoxi Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Xiaoxi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130234102","display_name":"Yuan Lu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Yuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5034516630","display_name":"Xinggao Liu","orcid":"https://orcid.org/0000-0002-0948-1942"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Xinggao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130232335","display_name":"Haoxuan Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Haoxuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5130242239","display_name":"Zhouchen Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Zhouchen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"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/T10028","display_name":"Topic Modeling","score":0.49799999594688416,"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/T10028","display_name":"Topic Modeling","score":0.49799999594688416,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.07919999957084656,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.04879999905824661,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/observational-study","display_name":"Observational study","score":0.7146000266075134},{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.6172000169754028},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5954999923706055},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.5350000262260437},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.4593000113964081},{"id":"https://openalex.org/keywords/observational-learning","display_name":"Observational learning","score":0.4481000006198883},{"id":"https://openalex.org/keywords/human-in-the-loop","display_name":"Human-in-the-loop","score":0.3653999865055084},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.36309999227523804},{"id":"https://openalex.org/keywords/annotation","display_name":"Annotation","score":0.33730000257492065}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.772599995136261},{"id":"https://openalex.org/C23131810","wikidata":"https://www.wikidata.org/wiki/Q818574","display_name":"Observational study","level":2,"score":0.7146000266075134},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.6172000169754028},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5971999764442444},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5954999923706055},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.5350000262260437},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4713999927043915},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.4593000113964081},{"id":"https://openalex.org/C102483320","wikidata":"https://www.wikidata.org/wiki/Q1143725","display_name":"Observational learning","level":3,"score":0.4481000006198883},{"id":"https://openalex.org/C2780626000","wikidata":"https://www.wikidata.org/wiki/Q5936775","display_name":"Human-in-the-loop","level":2,"score":0.3653999865055084},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.36309999227523804},{"id":"https://openalex.org/C2776321320","wikidata":"https://www.wikidata.org/wiki/Q857525","display_name":"Annotation","level":2,"score":0.33730000257492065},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.328900009393692},{"id":"https://openalex.org/C67712803","wikidata":"https://www.wikidata.org/wiki/Q7901853","display_name":"User modeling","level":3,"score":0.31929999589920044},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.30559998750686646},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.28859999775886536},{"id":"https://openalex.org/C158600405","wikidata":"https://www.wikidata.org/wiki/Q5054566","display_name":"Causal inference","level":2,"score":0.2872999906539917},{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.2858999967575073},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.28279998898506165},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.27730000019073486},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.27559998631477356},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.2711000144481659},{"id":"https://openalex.org/C186886427","wikidata":"https://www.wikidata.org/wiki/Q5441213","display_name":"Feedback loop","level":2,"score":0.26339998841285706},{"id":"https://openalex.org/C176743888","wikidata":"https://www.wikidata.org/wiki/Q862797","display_name":"Observational methods in psychology","level":3,"score":0.260699987411499},{"id":"https://openalex.org/C557471498","wikidata":"https://www.wikidata.org/wiki/Q554950","display_name":"Recommender system","level":2,"score":0.2522999942302704},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.2515000104904175}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.18736","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.18736","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.18736","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.18736","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Despite":[0],"the":[1,151,160,174],"success":[2],"of":[3,176],"reinforcement":[4],"learning":[5,41],"from":[6,24,80,132,216],"human":[7,25],"feedback":[8,21,47,70,86,96,180,221],"(RLHF)":[9],"in":[10,65],"aligning":[11],"language":[12],"models,":[13],"current":[14],"reward":[15,38,42,122,130,214],"modeling":[16,39,123,159],"heavily":[17],"relies":[18],"on":[19,97,227,236,242,247],"experimental":[20],"data":[22],"collected":[23],"annotators":[26],"under":[27,154],"controlled":[28],"and":[29,51,57,110,205,218,222,238],"costly":[30],"conditions.":[31],"In":[32],"this":[33,66],"work,":[34],"we":[35,117],"introduce":[36],"observational":[37,45,69,85,133,220],"--":[40,53,173,185,231],"models":[43,131],"with":[44],"user":[46,82,90,178,196],"(e.g.,":[48],"clicks,":[49],"copies,":[50],"upvotes)":[52],"as":[54],"a":[55,105,120,141,177,182,191,233,239],"scalable":[56],"cost-effective":[58],"alternative.":[59],"We":[60],"identify":[61],"two":[62],"fundamental":[63],"challenges":[64],"setting:":[67],"(1)":[68],"is":[71,87,147,245],"noisy":[72,217],"due":[73],"to":[74,127,150,186],"annotation":[75,161],"errors,":[76],"which":[77,103],"deviates":[78],"it":[79],"true":[81],"preference;":[83],"(2)":[84],"biased":[88,219],"by":[89,157],"preference,":[91],"where":[92],"users":[93],"preferentially":[94],"provide":[95],"responses":[98],"they":[99],"feel":[100],"strongly":[101],"about,":[102],"creats":[104],"distribution":[106],"shift":[107],"between":[108],"training":[109,188],"inference":[111],"data.":[112],"To":[113,135,165],"address":[114],"these":[115],"challenges,":[116],"propose":[118],"CausalRM,":[119],"causal-theoretic":[121],"framework":[124],"that":[125,146,194,209],"aims":[126],"learn":[128],"unbiased":[129],"feedback.":[134],"tackle":[136,166],"challenge":[137,167],"(1),":[138],"CausalRM":[139,169,210],"introduces":[140],"noise-aware":[142],"surrogate":[143],"loss":[144,153,192],"term":[145],"provably":[148],"equivalent":[149],"primal":[152],"noise-free":[155],"conditions":[156],"explicitly":[158],"error":[162],"generation":[163],"process.":[164],"(2),":[168],"uses":[170],"propensity":[171],"scores":[172],"probability":[175],"providing":[179],"for":[181],"given":[183],"response":[184],"reweight":[187],"samples,":[189],"yielding":[190],"function":[193],"eliminates":[195],"preference":[197],"bias.":[198],"Extensive":[199],"experiments":[200],"across":[201],"diverse":[202],"LLM":[203],"backbones":[204],"benchmark":[206],"datasets":[207],"validate":[208],"effectively":[211],"learns":[212],"accurate":[213],"signals":[215],"delivers":[223],"substantial":[224],"performance":[225],"improvements":[226],"downstream":[228],"RLHF":[229],"tasks":[230],"including":[232],"49.2%":[234],"gain":[235],"WildGuardMix":[237],"32.7%":[240],"improvement":[241],"HarmBench.":[243],"Code":[244],"available":[246],"our":[248],"project":[249],"website.":[250]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-03-21T00:00:00"}