{"id":"https://openalex.org/W4412889728","doi":"https://doi.org/10.18653/v1/2025.acl-long.1421","title":"Bypass Back-propagation: Optimization-based Structural Pruning for Large Language Models via Policy Gradient","display_name":"Bypass Back-propagation: Optimization-based Structural Pruning for Large Language Models via Policy Gradient","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412889728","doi":"https://doi.org/10.18653/v1/2025.acl-long.1421"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.acl-long.1421","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1421","pdf_url":"https://aclanthology.org/2025.acl-long.1421.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 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.acl-long.1421.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5011958968","display_name":"Yuan Gao","orcid":"https://orcid.org/0000-0002-2990-9205"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuan Gao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111233217","display_name":"Zujing Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zujing Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5018458399","display_name":"Weizhong Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Weizhong Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5060042752","display_name":"Bo Du","orcid":"https://orcid.org/0000-0002-0059-8458"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bo Du","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5073032922","display_name":"Gui-Song Xia","orcid":"https://orcid.org/0000-0001-7660-6090"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gui-Song Xia","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":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"29356","last_page":"29377"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9391000270843506,"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":0.9391000270843506,"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/pruning","display_name":"Pruning","score":0.6910853981971741},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6569011807441711},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4402506947517395},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4030114710330963},{"id":"https://openalex.org/keywords/mathematical-optimization","display_name":"Mathematical optimization","score":0.3359314203262329},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.15771430730819702}],"concepts":[{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.6910853981971741},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6569011807441711},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4402506947517395},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4030114710330963},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3359314203262329},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.15771430730819702},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0},{"id":"https://openalex.org/C6557445","wikidata":"https://www.wikidata.org/wiki/Q173113","display_name":"Agronomy","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.acl-long.1421","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1421","pdf_url":"https://aclanthology.org/2025.acl-long.1421.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 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.acl-long.1421","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1421","pdf_url":"https://aclanthology.org/2025.acl-long.1421.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 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G7875021329","display_name":null,"funder_award_id":"62325111","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320321885","display_name":"Science and Technology Commission of Shanghai Municipality","ror":"https://ror.org/03kt66j61"},{"id":"https://openalex.org/F4320322186","display_name":"Natural Science Foundation of Hubei Province","ror":null}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412889728.pdf","grobid_xml":"https://content.openalex.org/works/W4412889728.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Recent":[0],"Large-Language":[1],"Models":[2],"(LLMs)":[3],"pruning":[4,19,38,42,90,118],"methods":[5],"typically":[6],"operate":[7],"at":[8,168],"the":[9,13,41,51,54,62,65,73,77,95,150,156],"posttraining":[10],"phase":[11],"without":[12,108],"expensive":[14],"weight":[15],"finetuning,":[16],"however,":[17],"their":[18],"criteria":[20],"often":[21],"rely":[22],"on":[23,141],"heuristically":[24],"hand-crafted":[25],"metrics,":[26],"potentially":[27],"leading":[28],"to":[29,87],"suboptimal":[30],"performance.We":[31],"instead":[32],"propose":[33],"a":[34,45,132],"novel":[35],"optimizationbased":[36],"structural":[37],"that":[39],"learns":[40],"masks":[43],"in":[44,162],"probabilistic":[46],"space":[47],"directly":[48],"by":[49,81],"optimizing":[50],"loss":[52],"of":[53,76,159],"pruned":[55],"model.To":[56],"preserve":[57],"efficiency,":[58],"our":[59,110,136,160],"method":[60,111,134,161],"eliminates":[61],"back-propagation":[63],"through":[64],"LLM":[66,99],"per":[67],"se":[68],"during":[69],"optimization,":[70],"requiring":[71],"only":[72],"forward":[74],"pass":[75],"LLM.We":[78],"achieve":[79],"this":[80],"learning":[82],"an":[83],"underlying":[84],"Bernoulli":[85,96,137],"distribution":[86],"sample":[88],"binary":[89],"masks,":[91],"where":[92],"we":[93],"decouple":[94],"parameters":[97],"from":[98],"loss,":[100],"facilitating":[101],"efficient":[102],"optimization":[103],"via":[104],"policy":[105],"gradient":[106],"estimator":[107],"back-propagation.Thus,":[109],"can":[112],"1)":[113],"support":[114],"global":[115],"and":[116,127,146,152,164],"heterogeneous":[117],"(i.e.,":[119],"automatically":[120],"determine":[121],"different":[122,125],"redundancy":[123],"for":[124],"layers),":[126],"2)":[128],"optionally":[129],"initialize":[130],"with":[131],"metric-based":[133],"(for":[135],"distributions).Extensive":[138],"experiments":[139],"conducted":[140],"LLaMA,":[142],"LLaMA-2,":[143],"LLaMA-3,":[144],"Vicuna,":[145],"Mistral":[147],"models":[148],"using":[149],"C4":[151],"WikiText2":[153],"datasets":[154],"demonstrate":[155],"promising":[157],"performance":[158],"efficiency":[163],"effectiveness.Code":[165],"is":[166],"available":[167],"https://github.com/":[169],"ethanygao/backprop-free_LLM_pruning.":[170]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}