{"id":"https://openalex.org/W4415330907","doi":"https://doi.org/10.48550/arxiv.2509.22691","title":"Sequential Token Merging: Revisiting Hidden States","display_name":"Sequential Token Merging: Revisiting Hidden States","publication_year":2025,"publication_date":"2025-09-19","ids":{"openalex":"https://openalex.org/W4415330907","doi":"https://doi.org/10.48550/arxiv.2509.22691"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2509.22691","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2509.22691","pdf_url":"https://arxiv.org/pdf/2509.22691","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2509.22691","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Wen, Yan","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wen, Yan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Ye, Peng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ye, Peng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100889409","display_name":"Lin Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Lin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5087869694","display_name":"Baopu Li","orcid":"https://orcid.org/0000-0002-9032-3991","observed_orcids":["https://orcid.org/0000-0002-9032-3991"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Baopu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5056327596","display_name":"Jiakang Yuan","orcid":"https://orcid.org/0009-0006-6651-4356","observed_orcids":["https://orcid.org/0009-0006-6651-4356"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuan, Jiakang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Yang, Yaoxin","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Yaoxin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":null,"display_name":"Chen, Tao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Tao","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/T10036","display_name":"Advanced Neural Network Applications","score":0.9922000169754028,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.9922000169754028,"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/T11605","display_name":"Visual Attention and Saliency Detection","score":0.00279999990016222,"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":0.0013000000035390258,"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/vision-mamba","display_name":"Vision Mamba","score":0.8700000047683716},{"id":"https://openalex.org/keywords/token-merging","display_name":"token merging","score":0.652999997138977},{"id":"https://openalex.org/keywords/hidden-states","display_name":"hidden states","score":0.7310000061988831},{"id":"https://openalex.org/keywords/selective-scanning","display_name":"selective scanning","score":0.722000002861023},{"id":"https://openalex.org/keywords/token-reduction","display_name":"token reduction","score":0.6129999756813049},{"id":"https://openalex.org/keywords/state-space-models","display_name":"state-space model","score":0.7929999828338623},{"id":"https://openalex.org/keywords/image-compression","display_name":"image compression","score":0.3529999852180481}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.8418999910354614},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7390000224113464},{"id":"https://openalex.org/C18664526","wikidata":"https://www.wikidata.org/wiki/Q7651187","display_name":"Suzuki-Kasami algorithm","level":3,"score":0.5212000012397766},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.5031999945640564},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.4652999937534332},{"id":"https://openalex.org/C129844170","wikidata":"https://www.wikidata.org/wiki/Q41299","display_name":"Quadratic equation","level":2,"score":0.44530001282691956},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4343000054359436},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.43309998512268066},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.375},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.36640000343322754},{"id":"https://openalex.org/C115067241","wikidata":"https://www.wikidata.org/wiki/Q1639854","display_name":"Token passing","level":3,"score":0.36559998989105225},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.35519999265670776},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.351500004529953},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.34389999508857727},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.32280001044273376},{"id":"https://openalex.org/C89611455","wikidata":"https://www.wikidata.org/wiki/Q6804646","display_name":"Mechanism (biology)","level":2,"score":0.3165000081062317},{"id":"https://openalex.org/C2988416141","wikidata":"https://www.wikidata.org/wiki/Q6031139","display_name":"Information loss","level":2,"score":0.2639000117778778},{"id":"https://openalex.org/C81845259","wikidata":"https://www.wikidata.org/wiki/Q290117","display_name":"Quadratic programming","level":2,"score":0.262800008058548},{"id":"https://openalex.org/C2779136372","wikidata":"https://www.wikidata.org/wiki/Q10283002","display_name":"Information flow","level":2,"score":0.2535000145435333},{"id":"https://openalex.org/C2983524802","wikidata":"https://www.wikidata.org/wiki/Q41299","display_name":"Quadratic model","level":3,"score":0.2529999911785126},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2515999972820282},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.25060001015663147}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2509.22691","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2509.22691","pdf_url":"https://arxiv.org/pdf/2509.22691","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2509.22691","doi":"https://doi.org/10.48550/arxiv.2509.22691","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2509.22691","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,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2509.22691","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2509.22691","pdf_url":"https://arxiv.org/pdf/2509.22691","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4405562057","https://openalex.org/W4414992083","https://openalex.org/W4416281203","https://openalex.org/W7165397691","https://openalex.org/W7133317294","https://openalex.org/W7204951511","https://openalex.org/W4416976402","https://openalex.org/W7045370771","https://openalex.org/W4413146088","https://openalex.org/W4415798456"],"abstract_inverted_index":{"Vision":[0],"Mambas":[1],"(ViMs)":[2],"achieve":[3],"remarkable":[4],"success":[5],"with":[6,18,137],"sub-quadratic":[7],"complexity,":[8,139],"but":[9],"their":[10],"efficiency":[11,136],"remains":[12],"constrained":[13],"by":[14],"quadratic":[15],"token":[16,25,121],"scaling":[17],"image":[19],"resolution.":[20],"While":[21],"existing":[22],"methods":[23],"address":[24],"redundancy,":[26],"they":[27],"overlook":[28],"ViMs'":[29],"intrinsic":[30],"Limited":[31],"Directional":[32],"Sequential":[33,65],"Dependence":[34],"(LDSD)":[35],"-":[36],"a":[37],"critical":[38],"information":[39,54],"flow":[40],"mechanism":[41],"revealed":[42],"in":[43,56],"our":[44],"analysis.":[45],"We":[46],"further":[47],"identify":[48],"Mamba's":[49,100],"selective":[50],"scan":[51],"enables":[52],"gradual":[53],"aggregation":[55],"hidden":[57,91],"states.":[58],"Based":[59],"on":[60],"these":[61],"insights,":[62],"we":[63],"propose":[64],"Token":[66],"Merging":[67],"(STM),":[68],"featuring:":[69],"1)":[70],"Bidirectional":[71],"nearest":[72],"neighbor":[73],"merging":[74],"to":[75,88,104],"preserve":[76],"sequential":[77],"dependencies":[78],"through":[79],"symmetric":[80],"spatial":[81],"aggregation,":[82],"and":[83,123],"2)":[84],"Hidden":[85],"states":[86,92],"protection":[87],"stabilize":[89],"the":[90,94],"around":[93],"class":[95],"token.":[96],"STM":[97],"strategically":[98],"leverages":[99],"layer-wise":[101],"loss":[102],"convergence":[103],"convert":[105],"temporal":[106],"forgetfulness":[107],"into":[108,144],"stability.":[109],"Experiments":[110],"demonstrate":[111],"STM's":[112],"superiority:":[113],"1.0%":[114],"accuracy":[115],"drop":[116],"for":[117,127],"ViM-Ti":[118],"at":[119,129],"20%":[120],"reduction,":[122],"only":[124],"1.4%":[125],"degradation":[126],"ViM-S":[128],"40%":[130],"reduction.":[131],"Our":[132],"method":[133],"achieves":[134],"state-of-the-art":[135],"minimal":[138],"while":[140],"providing":[141],"new":[142],"insights":[143],"state-space":[145],"model":[146],"dynamics.":[147],"Codes":[148],"will":[149],"be":[150],"released":[151],"soon.":[152]},"counts_by_year":[],"updated_date":"2026-10-14T16:39:01.820549","created_date":"2025-10-19T00:00:00"}