{"id":"https://openalex.org/W4416035602","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.1338","title":"GRASP: Replace Redundant Layers with Adaptive Singular Parameters for Efficient Model Compression","display_name":"GRASP: Replace Redundant Layers with Adaptive Singular Parameters for Efficient Model Compression","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416035602","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.1338"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.emnlp-main.1338","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1338","pdf_url":"https://aclanthology.org/2025.emnlp-main.1338.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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.emnlp-main.1338.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5016104023","display_name":"Kainan Liu","orcid":"https://orcid.org/0009-0005-6005-3619"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Kainan Liu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100619208","display_name":"Yong Zhang","orcid":"https://orcid.org/0000-0003-4909-8996"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yong Zhang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5065576761","display_name":"Ning Cheng","orcid":"https://orcid.org/0000-0003-0909-5328"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ning Cheng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101511899","display_name":"Zhitao Li","orcid":"https://orcid.org/0000-0001-9967-1303"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhitao Li","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100702012","display_name":"Shaojun Wang","orcid":"https://orcid.org/0000-0002-3680-9464"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shaojun Wang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5102172482","display_name":"Jing Xiao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jing Xiao","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":6,"corresponding_author_ids":["https://openalex.org/A5016104023"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.36112262,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"26344","last_page":"26359"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10904","display_name":"Embedded Systems Design Techniques","score":0.2378000020980835,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T10904","display_name":"Embedded Systems Design Techniques","score":0.2378000020980835,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T10901","display_name":"Advanced Data Compression Techniques","score":0.15399999916553497,"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/T12292","display_name":"Graph Theory and Algorithms","score":0.1136000007390976,"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/compression","display_name":"Compression (physics)","score":0.4961000084877014},{"id":"https://openalex.org/keywords/data-compression","display_name":"Data compression","score":0.3517000079154968},{"id":"https://openalex.org/keywords/control-theory","display_name":"Control theory (sociology)","score":0.29330000281333923},{"id":"https://openalex.org/keywords/redundancy","display_name":"Redundancy (engineering)","score":0.2770000100135803}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5228999853134155},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.4961000084877014},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4377000033855438},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.3517000079154968},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3319999873638153},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.29330000281333923},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.28459998965263367},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.2770000100135803},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.2685000002384186},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.2565999925136566}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.18653/v1/2025.emnlp-main.1338","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1338","pdf_url":"https://aclanthology.org/2025.emnlp-main.1338.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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},{"id":"pmh:doi:10.48550/arxiv.2501.00339","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"publisher-specific-oa","license_id":"https://openalex.org/licenses/publisher-specific-oa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.emnlp-main.1338","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1338","pdf_url":"https://aclanthology.org/2025.emnlp-main.1338.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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416035602.pdf","grobid_xml":"https://content.openalex.org/works/W4416035602.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Recent":[0],"studies":[1],"have":[2],"demonstrated":[3],"that":[4,55,109],"many":[5],"layers":[6,21,86],"are":[7],"functionally":[8],"redundant":[9,85],"in":[10,36],"large":[11],"language":[12],"models":[13],"(LLMs),":[14],"enabling":[15],"model":[16],"compression":[17,53,97,114,125],"by":[18,59],"removing":[19],"these":[20],"to":[22,76],"reduce":[23],"inference":[24],"cost.While":[25],"such":[26],"approaches":[27],"can":[28],"improve":[29],"efficiency,":[30],"indiscriminate":[31],"layer":[32,65],"pruning":[33],"often":[34],"results":[35],"significant":[37],"performance":[38,101,122],"degradation.In":[39],"this":[40,57],"paper,":[41],"we":[42],"propose":[43],"GRASP":[44,67,94,110],"(Gradient-based":[45],"Retention":[46],"of":[47,92,118],"Adaptive":[48],"Singular":[49],"Parameters),":[50],"a":[51,72,89],"novel":[52],"framework":[54],"mitigates":[56],"issue":[58],"preserving":[60],"sensitivity-aware":[61],"singular":[62,82],"values.Unlike":[63],"direct":[64],"pruning,":[66],"leverages":[68],"gradient-based":[69],"attribution":[70],"on":[71],"small":[73],"calibration":[74],"dataset":[75],"adaptively":[77],"identify":[78],"and":[79],"retain":[80],"critical":[81],"components.By":[83],"replacing":[84],"with":[87,102],"only":[88],"minimal":[90,103],"set":[91],"parameters,":[93],"achieves":[95],"efficient":[96],"while":[98],"maintaining":[99],"strong":[100],"overhead.Experiments":[104],"across":[105],"multiple":[106],"LLMs":[107],"show":[108],"consistently":[111],"outperforms":[112],"existing":[113],"methods,":[115],"achieving":[116],"90%":[117],"the":[119],"original":[120],"model's":[121],"under":[123],"20%":[124],"ratio.The":[126],"source":[127],"code":[128],"is":[129],"available":[130],"at":[131],"https://github.com/LyoAI/GRASP.":[132]},"counts_by_year":[],"updated_date":"2026-04-04T16:13:02.066488","created_date":"2025-11-08T00:00:00"}