{"id":"https://openalex.org/W7161301042","doi":"https://doi.org/10.48550/arxiv.2605.14448","title":"Think When Needed: Adaptive Reasoning-Driven Multimodal Embeddings with a Dual-LoRA Architecture","display_name":"Think When Needed: Adaptive Reasoning-Driven Multimodal Embeddings with a Dual-LoRA Architecture","publication_year":2026,"publication_date":"2026-05-14","ids":{"openalex":"https://openalex.org/W7161301042","doi":"https://doi.org/10.48550/arxiv.2605.14448"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.14448","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.14448","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,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.14448","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136249367","display_name":"Longxiang Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Longxiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102266633","display_name":"Weilong Dai","orcid":"https://orcid.org/0009-0006-9288-932X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dai, Weilong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136196845","display_name":"Guanghao Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Guanghao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136212056","display_name":"Hao Jiang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiang, Hao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5059615376","display_name":"Pipei Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Pipei","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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.4090000092983246,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.4090000092983246,"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/T10028","display_name":"Topic Modeling","score":0.3691999912261963,"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/T11273","display_name":"Advanced Graph Neural Networks","score":0.10080000013113022,"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/embedding","display_name":"Embedding","score":0.7659000158309937},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6359999775886536},{"id":"https://openalex.org/keywords/discriminative-model","display_name":"Discriminative model","score":0.5906999707221985},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.5447999835014343},{"id":"https://openalex.org/keywords/semantic-reasoner","display_name":"Semantic reasoner","score":0.483599990606308},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.47269999980926514},{"id":"https://openalex.org/keywords/quality","display_name":"Quality (philosophy)","score":0.40619999170303345},{"id":"https://openalex.org/keywords/generative-grammar","display_name":"Generative grammar","score":0.4050000011920929},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.39430001378059387}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7957000136375427},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.7659000158309937},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6359999775886536},{"id":"https://openalex.org/C97931131","wikidata":"https://www.wikidata.org/wiki/Q5282087","display_name":"Discriminative model","level":2,"score":0.5906999707221985},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5648000240325928},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.5447999835014343},{"id":"https://openalex.org/C9616225","wikidata":"https://www.wikidata.org/wiki/Q3929429","display_name":"Semantic reasoner","level":2,"score":0.483599990606308},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.47269999980926514},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.40619999170303345},{"id":"https://openalex.org/C39890363","wikidata":"https://www.wikidata.org/wiki/Q36108","display_name":"Generative grammar","level":2,"score":0.4050000011920929},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.39430001378059387},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3792000114917755},{"id":"https://openalex.org/C103278499","wikidata":"https://www.wikidata.org/wiki/Q254465","display_name":"Similarity (geometry)","level":3,"score":0.3758000135421753},{"id":"https://openalex.org/C49937458","wikidata":"https://www.wikidata.org/wiki/Q2599292","display_name":"Probabilistic logic","level":2,"score":0.3352000117301941},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.31839999556541443},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.3149000108242035},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.3052999973297119},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.30160000920295715},{"id":"https://openalex.org/C113843644","wikidata":"https://www.wikidata.org/wiki/Q901882","display_name":"Interface (matter)","level":4,"score":0.2971000075340271},{"id":"https://openalex.org/C18555067","wikidata":"https://www.wikidata.org/wiki/Q8375051","display_name":"Joint (building)","level":2,"score":0.28760001063346863},{"id":"https://openalex.org/C20162079","wikidata":"https://www.wikidata.org/wiki/Q1151406","display_name":"Case-based reasoning","level":2,"score":0.27000001072883606},{"id":"https://openalex.org/C167966045","wikidata":"https://www.wikidata.org/wiki/Q5532625","display_name":"Generative model","level":3,"score":0.26969999074935913},{"id":"https://openalex.org/C123657996","wikidata":"https://www.wikidata.org/wiki/Q12271","display_name":"Architecture","level":2,"score":0.2648000121116638},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.2646999955177307},{"id":"https://openalex.org/C2781020372","wikidata":"https://www.wikidata.org/wiki/Q533093","display_name":"On the fly","level":2,"score":0.2556999921798706},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.25130000710487366}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.14448","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.14448","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,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.14448","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.14448","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,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"display_name":"Reduced inequalities","id":"https://metadata.un.org/sdg/10","score":0.6740846633911133}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Multimodal":[0],"large":[1],"language":[2],"models":[3],"(MLLMs)":[4],"have":[5],"emerged":[6],"as":[7],"a":[8,88,98,108,130,141],"powerful":[9],"backbone":[10,205],"for":[11,53,60],"multimodal":[12,90],"embeddings.":[13],"Recent":[14],"methods":[15],"introduce":[16],"chain-of-thought":[17],"(CoT)":[18],"reasoning":[19,70,103,155,211],"into":[20],"the":[21,74,178,204,215],"embedding":[22,91,105,186],"pipeline":[23],"to":[24,107,117,129,145,150,156,170,203,208,214],"improve":[25,162],"retrieval":[26,163],"quality,":[27],"but":[28],"remain":[29],"costly":[30],"in":[31],"both":[32],"model":[33],"size":[34],"and":[35,43,49,68,104,160,206],"inference":[36,158],"cost.":[37],"They":[38],"typically":[39],"employ":[40],"separate":[41],"reasoner":[42],"embedder":[44],"with":[45,93],"substantial":[46],"parameter":[47],"overhead,":[48],"generate":[50,151],"CoT":[51,172],"indiscriminately":[52],"every":[54],"input.":[55],"However,":[56],"we":[57,82],"observe":[58],"that":[59,101],"simple":[61],"inputs,":[62],"discriminative":[63],"embeddings":[64],"already":[65],"perform":[66],"well,":[67],"redundant":[69],"can":[71],"even":[72,161],"mislead":[73],"model,":[75],"degrading":[76],"performance.":[77],"To":[78],"address":[79],"these":[80],"limitations,":[81],"propose":[83],"Think":[84],"When":[85],"Needed":[86],"(TWN),":[87],"unified":[89],"framework":[92],"adaptive":[94,137],"reasoning.":[95],"TWN":[96,183],"introduces":[97],"dual-LoRA":[99],"architecture":[100],"attaches":[102],"adapters":[106],"shared":[109],"frozen":[110],"backbone,":[111],"detaching":[112],"gradients":[113],"at":[114],"their":[115],"interface":[116],"mitigate":[118],"gradient":[119],"conflicts":[120],"introduced":[121],"by":[122],"joint":[123],"optimization":[124],"while":[125,188],"keeping":[126],"parameters":[127,201],"close":[128],"single":[131],"model.":[132],"Building":[133],"on":[134],"this,":[135],"an":[136],"think":[138],"mechanism":[139],"uses":[140],"self-supervised":[142],"routing":[143],"gate":[144],"decide":[146],"per":[147],"input":[148],"whether":[149],"CoT,":[152],"skipping":[153],"unnecessary":[154],"reduce":[157],"overhead":[159],"quality.":[164],"We":[165],"further":[166],"explore":[167],"embedding-guided":[168],"RL":[169],"optimize":[171],"quality":[173,187],"beyond":[174],"supervised":[175],"training.":[176],"On":[177],"78":[179],"tasks":[180],"of":[181],"MMEB-V2,":[182],"achieves":[184],"state-of-the-art":[185],"being":[189],"substantially":[190],"more":[191],"efficient":[192],"than":[193],"existing":[194],"generative":[195,217],"methods,":[196],"requiring":[197],"only":[198],"3-5%":[199],"additional":[200],"relative":[202],"up":[207],"50%":[209],"fewer":[210],"tokens":[212],"compared":[213],"full":[216],"mode.":[218]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-16T00:00:00"}