{"id":"https://openalex.org/W7166835578","doi":"https://doi.org/10.18653/v1/2026.acl-short.45","title":"Taming Extreme Tokens: Covariance-Aware GRPO with Gaussian-Kernel Advantage Reweighting","display_name":"Taming Extreme Tokens: Covariance-Aware GRPO with Gaussian-Kernel Advantage Reweighting","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166835578","doi":"https://doi.org/10.18653/v1/2026.acl-short.45"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.acl-short.45","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-short.45","pdf_url":"https://aclanthology.org/2026.acl-short.45.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.acl-short.45.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139794326","display_name":"Cheng Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cheng Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139809413","display_name":"Qin Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qin Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139772315","display_name":"Wenxuan Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wenxuan Zhou","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139789274","display_name":"Muhao Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Muhao Chen","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":0.82527397,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"540","last_page":"546"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.36039999127388,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.36039999127388,"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/T12814","display_name":"Gaussian Processes and Bayesian Inference","score":0.11299999803304672,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.0714000016450882,"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/set","display_name":"Set (abstract data type)","score":0.2840999960899353},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.25999999046325684},{"id":"https://openalex.org/keywords/bayesian-probability","display_name":"Bayesian probability","score":0.24650000035762787},{"id":"https://openalex.org/keywords/sequence","display_name":"Sequence (biology)","score":0.24539999663829803},{"id":"https://openalex.org/keywords/identification","display_name":"Identification (biology)","score":0.24529999494552612}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4521999955177307},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.2840999960899353},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.25999999046325684},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2524000108242035},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2475000023841858},{"id":"https://openalex.org/C107673813","wikidata":"https://www.wikidata.org/wiki/Q812534","display_name":"Bayesian probability","level":2,"score":0.24650000035762787},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.24539999663829803},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.24529999494552612},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.24220000207424164},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.2379000037908554}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.acl-short.45","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-short.45","pdf_url":"https://aclanthology.org/2026.acl-short.45.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.acl-short.45","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-short.45","pdf_url":"https://aclanthology.org/2026.acl-short.45.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G1795802330","display_name":"CICI: UCSS: Enabling a Safe and Directive Multi-modal Foundation Model Ecosystem for Food Science Research","funder_award_id":"2531126","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G4706632707","display_name":"Proto-OKN Theme 1: Knowledge Graph Construction for Resilient, Trustworthy, and Secure Software Supply Chains","funder_award_id":"2333736","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":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166835578.pdf","grobid_xml":"https://content.openalex.org/works/W7166835578.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Group":[0],"Relative":[1],"Policy":[2],"Optimization":[3],"(GRPO)":[4],"has":[5],"emerged":[6],"as":[7,108],"a":[8,60,72],"promising":[9],"approach":[10,75,94],"for":[11],"improving":[12],"the":[13,25,39,49,78],"reasoning":[14,99],"capabilities":[15],"of":[16],"large":[17],"language":[18],"models.However,":[19],"it":[20],"struggles":[21],"to":[22],"effectively":[23,105],"balance":[24],"tradeoff":[26],"between":[27,51],"exploration":[28],"and":[29,54,104],"exploitation":[30],"during":[31],"training,":[32],"often":[33],"resulting":[34],"in":[35,44],"suboptimal":[36],"performance.Motivated":[37],"by":[38,48,81],"theoretical":[40],"insight":[41],"that":[42,65,92],"changes":[43],"entropy":[45,107],"are":[46],"governed":[47],"covariance":[50],"token":[52],"probabilities":[53],"their":[55],"corresponding":[56],"advantages,":[57],"we":[58],"propose":[59],"hyperparameter-free,":[61],"covariance-weighted":[62],"optimization":[63],"method":[64],"dynamically":[66],"down-weights":[67],"extreme":[68],"token-level":[69],"updates":[70],"via":[71],"Gaussian":[73],"kernel.This":[74],"automatically":[76],"reduces":[77],"instability":[79],"caused":[80],"exploration-exploitation":[82],"trade-off":[83],"while":[84],"preserving":[85],"informative":[86],"learning":[87],"signals.Extensive":[88],"empirical":[89],"evaluations":[90],"show":[91],"our":[93],"improves":[95],"downstream":[96],"performance":[97],"across":[98],"benchmarks":[100],"compared":[101],"with":[102],"GRPO,":[103],"stablizes":[106],"training":[109],"progresses.":[110]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}