{"id":"https://openalex.org/W7166801979","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1568","title":"The Best of Both Worlds: Combining Parallel and Sequential Inference Scaling via Aggregation Fine-Tuning","display_name":"The Best of Both Worlds: Combining Parallel and Sequential Inference Scaling via Aggregation Fine-Tuning","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166801979","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1568"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.1568","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1568","pdf_url":"https://aclanthology.org/2026.findings-acl.1568.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.1568.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5032649444","display_name":"Yafu Li","orcid":"https://orcid.org/0000-0002-7895-9997"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yafu Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139831445","display_name":"Zhilin Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhilin Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139834316","display_name":"Tingchen Fu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tingchen Fu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102542015","display_name":"Ganqu Cui","orcid":"https://orcid.org/0000-0001-6385-8547"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ganqu Cui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139740690","display_name":"Sen Yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sen Yang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139851917","display_name":"Yu Cheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu Cheng","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":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.85743777,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"31369","last_page":"31389"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.1632000058889389,"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"}},"topics":[{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.1632000058889389,"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"}},{"id":"https://openalex.org/T10057","display_name":"Face and Expression Recognition","score":0.08940000087022781,"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"}},{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.07530000060796738,"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/inference","display_name":"Inference","score":0.4293000102043152},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.42559999227523804},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.28700000047683716},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.2685000002384186},{"id":"https://openalex.org/keywords/sequence","display_name":"Sequence (biology)","score":0.25850000977516174}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5663999915122986},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5385000109672546},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4293000102043152},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.42559999227523804},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.37689998745918274},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.33009999990463257},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.28700000047683716},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2685000002384186},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.25850000977516174},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.25760000944137573},{"id":"https://openalex.org/C86426650","wikidata":"https://www.wikidata.org/wiki/Q7452504","display_name":"Sequential estimation","level":2,"score":0.2556000053882599}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.1568","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1568","pdf_url":"https://aclanthology.org/2026.findings-acl.1568.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.1568","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1568","pdf_url":"https://aclanthology.org/2026.findings-acl.1568.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166801979.pdf","grobid_xml":"https://content.openalex.org/works/W7166801979.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Scaling":[0],"data":[1],"and":[2,75,147],"model":[3,47],"size":[4],"has":[5],"been":[6],"proven":[7],"effective":[8],"for":[9,113],"boosting":[10],"the":[11,46,111,150],"performance":[12],"of":[13],"large":[14],"language":[15],"models.In":[16],"addition":[17],"to":[18,49,55],"training-time":[19],"scaling,":[20,105],"recent":[21],"studies":[22],"have":[23],"revealed":[24],"that":[25,72,97],"increasing":[26],"test-time":[27,104],"computational":[28],"resources":[29],"can":[30],"further":[31],"improve":[32],"performance.In":[33],"this":[34],"work,":[35],"we":[36,67],"introduce":[37],"Aggregation":[38],"Fine-Tuning":[39],"(AFT),":[40],"a":[41,59,69,89,128,157],"supervised":[42],"fine-tuning":[43],"paradigm":[44],"where":[45],"learns":[48],"synthesize":[50],"multiple":[51],"draft":[52],"responses,":[53],"referred":[54],"as":[56,88,141],"proposals,":[57,77],"into":[58],"single,":[60],"refined":[61],"answer,":[62],"termed":[63],"aggregation.At":[64],"inference":[65],"time,":[66],"apply":[68],"propose-and-aggregate":[70],"strategy":[71],"iteratively":[73],"generates":[74],"aggregates":[76],"effectively":[78],"scaling":[79],"inferencetime":[80],"computation":[81,155],"without":[82],"relying":[83],"on":[84,133],"external":[85,114],"guidance":[86],"such":[87,140],"reward":[90,115],"model.Empirical":[91],"results":[92],"across":[93],"benchmark":[94],"datasets":[95],"demonstrate":[96],"AFT-trained":[98],"models":[99],"achieve":[100],"substantial":[101],"gains":[102],"with":[103,123],"outperforming":[106],"best-of-N":[107],"baselines":[108],"while":[109],"eliminating":[110],"need":[112],"signals.Notably,":[116],"an":[117],"AFT":[118],"model,":[119],"fine-tuned":[120],"from":[121],"Llama3.1-8B-Base":[122],"only":[124],"64k":[125],"data,":[126],"achieves":[127],"41.3%":[129],"LC":[130],"win":[131],"rate":[132],"AlpacaEval":[134],"2,":[135],"surpassing":[136],"significantly":[137],"larger":[138],"LLMs":[139],"Llama3.1-405B-Instructand":[142],"GPT-4.By":[143],"combining":[144],"sequential":[145],"refinement":[146],"parallel":[148],"sampling,":[149],"proposeand-aggregate":[151],"framework":[152],"scales":[153],"inference-time":[154],"in":[156],"flexible":[158],"manner.":[159]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}