{"id":"https://openalex.org/W7162674349","doi":"https://doi.org/10.48550/arxiv.2605.28069","title":"ZipRL: Adaptive Multi-Turn Context Compression with Hindsight Response Replay","display_name":"ZipRL: Adaptive Multi-Turn Context Compression with Hindsight Response Replay","publication_year":2026,"publication_date":"2026-05-27","ids":{"openalex":"https://openalex.org/W7162674349","doi":"https://doi.org/10.48550/arxiv.2605.28069"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.28069","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28069","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,"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.2605.28069","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136500820","display_name":"Zhexin Hu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Zhexin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137298100","display_name":"Li Wang","orcid":"https://orcid.org/0000-0002-9615-1879"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137221081","display_name":"Xiaohan Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Xiaohan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137296668","display_name":"Jiajun Chai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chai, Jiajun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137208991","display_name":"Xiaojun Guo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Guo, Xiaojun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137205465","display_name":"Wei Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Wei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137267789","display_name":"Guojun Yin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yin, Guojun","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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.22840000689029694,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.22840000689029694,"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/T10028","display_name":"Topic Modeling","score":0.12759999930858612,"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/T13702","display_name":"Machine Learning in Healthcare","score":0.10949999839067459,"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/robustness","display_name":"Robustness (evolution)","score":0.5546000003814697},{"id":"https://openalex.org/keywords/data-compression","display_name":"Data compression","score":0.5461000204086304},{"id":"https://openalex.org/keywords/compression","display_name":"Compression (physics)","score":0.5264000296592712},{"id":"https://openalex.org/keywords/hindsight-bias","display_name":"Hindsight bias","score":0.5170000195503235},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.5070000290870667},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.4406000077724457},{"id":"https://openalex.org/keywords/extrapolation","display_name":"Extrapolation","score":0.3617999851703644}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.77920001745224},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.5546000003814697},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.5461000204086304},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.5264000296592712},{"id":"https://openalex.org/C10347200","wikidata":"https://www.wikidata.org/wiki/Q1960297","display_name":"Hindsight bias","level":2,"score":0.5170000195503235},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5070000290870667},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.45010000467300415},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.4406000077724457},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3831999897956848},{"id":"https://openalex.org/C132459708","wikidata":"https://www.wikidata.org/wiki/Q744069","display_name":"Extrapolation","level":2,"score":0.3617999851703644},{"id":"https://openalex.org/C94835093","wikidata":"https://www.wikidata.org/wiki/Q3113333","display_name":"Data compression ratio","level":5,"score":0.3441999852657318},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.34279999136924744},{"id":"https://openalex.org/C165021410","wikidata":"https://www.wikidata.org/wiki/Q55564","display_name":"Lossy compression","level":2,"score":0.3368000090122223},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.3100999891757965},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.28349998593330383},{"id":"https://openalex.org/C183322885","wikidata":"https://www.wikidata.org/wiki/Q17007702","display_name":"Context model","level":3,"score":0.2728999853134155},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.2621999979019165}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.28069","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28069","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,"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.2605.28069","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28069","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,"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":{"Adaptive":[0],"context":[1],"compression":[2,18,56,70],"is":[3],"vital":[4],"for":[5,59,72,108],"scaling":[6],"Large":[7],"Language":[8],"Models":[9],"(LLMs)":[10],"to":[11,31,43,86],"complex,":[12],"multi-turn":[13],"agent":[14,136],"tasks.":[15],"However,":[16],"rule-based":[17],"methods":[19],"may":[20],"discard":[21],"task-critical":[22],"nuances,":[23],"while":[24,152],"Reinforcement":[25,60],"Learning":[26,61],"(RL)":[27],"approaches":[28,142],"usually":[29],"struggle":[30],"balance":[32],"information":[33,75],"retention":[34],"and":[35,110,124,145,149,157],"token":[36,155],"efficiency":[37,156],"under":[38,159],"the":[39,128],"sparse":[40],"rewards":[41],"inherent":[42],"long-horizon":[44],"workflows.":[45],"To":[46],"bridge":[47],"this":[48],"gap,":[49],"we":[50,94],"propose":[51],"ZipRL,":[52],"a":[53,68,83],"novel":[54],"adaptive":[55],"framework":[57],"tailored":[58],"from":[62],"Verifiable":[63],"Rewards":[64],"(RLVR).":[65],"ZipRL":[66,104,139],"features":[67],"multi-granularity":[69],"mechanism":[71],"active,":[73],"non-uniform":[74],"reduction,":[76],"coupled":[77],"with":[78],"Hindsight":[79],"Response":[80],"Replay":[81],"(HRR),":[82],"technique":[84],"designed":[85],"densify":[87],"training":[88],"signals":[89],"during":[90],"RLVR":[91],"optimization.":[92],"Theoretically,":[93],"prove":[95],"ZipRL's":[96],"superior":[97],"task-relevant":[98],"utility":[99],"over":[100],"uniform":[101],"methods.":[102],"Concretely,":[103],"utilizes":[105],"coarse-to-fine":[106],"prompts":[107],"macro-compression":[109],"incorporates":[111],"HRR":[112],"into":[113],"GRPO":[114],"via":[115],"generalized":[116],"advantage":[117],"reshaping.":[118],"Multiple":[119],"models":[120],"of":[121,130],"varying":[122],"versions":[123],"parameter":[125],"scales":[126],"validate":[127],"effectiveness":[129],"our":[131],"approach.":[132],"Benchmarks":[133],"on":[134],"five":[135],"tasks":[137],"show":[138],"outperforms":[140],"state-of-the-art":[141],"by":[143],"27.9%":[144],"34.7%":[146],"across":[147],"Qwen3-4B":[148],"Qwen3-8B":[150],"models,":[151],"maintaining":[153],"exceptional":[154],"robustness":[158],"extreme":[160],"256-turn":[161],"extrapolation":[162],"stress":[163],"tests.":[164]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-29T00:00:00"}