{"id":"https://openalex.org/W7140329817","doi":"https://doi.org/10.48550/arxiv.2603.23483","title":"SpecEyes: Accelerating Agentic Multimodal LLMs via Speculative Perception and Planning","display_name":"SpecEyes: Accelerating Agentic Multimodal LLMs via Speculative Perception and Planning","publication_year":2026,"publication_date":"2026-03-24","ids":{"openalex":"https://openalex.org/W7140329817","doi":"https://doi.org/10.48550/arxiv.2603.23483"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.23483","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.23483","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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.2603.23483","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5130619363","display_name":"Haoyu Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Haoyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130574673","display_name":"Jinfa Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Jinfa","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130605468","display_name":"Zhongwei Wan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wan, Zhongwei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054226277","display_name":"Xiawu Zheng","orcid":"https://orcid.org/0000-0002-6855-5403"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng, Xiawu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130597579","display_name":"Rongrong Ji","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ji, Rongrong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5130584445","display_name":"Jiebo Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luo, Jiebo","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":0.798799991607666,"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":0.798799991607666,"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":0.053300000727176666,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.016100000590085983,"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/stateless-protocol","display_name":"Stateless protocol","score":0.6065999865531921},{"id":"https://openalex.org/keywords/exploit","display_name":"Exploit","score":0.5511000156402588},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.5170999765396118},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.4489000141620636},{"id":"https://openalex.org/keywords/oracle","display_name":"Oracle","score":0.43849998712539673},{"id":"https://openalex.org/keywords/perception","display_name":"Perception","score":0.42640000581741333},{"id":"https://openalex.org/keywords/surprise","display_name":"Surprise","score":0.4185999929904938},{"id":"https://openalex.org/keywords/stateful-firewall","display_name":"Stateful firewall","score":0.41440001130104065},{"id":"https://openalex.org/keywords/concurrency","display_name":"Concurrency","score":0.3869999945163727},{"id":"https://openalex.org/keywords/perspective","display_name":"Perspective (graphical)","score":0.375}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8177000284194946},{"id":"https://openalex.org/C103613024","wikidata":"https://www.wikidata.org/wiki/Q230924","display_name":"Stateless protocol","level":3,"score":0.6065999865531921},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.5511000156402588},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.5170999765396118},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.4489000141620636},{"id":"https://openalex.org/C55166926","wikidata":"https://www.wikidata.org/wiki/Q2892946","display_name":"Oracle","level":2,"score":0.43849998712539673},{"id":"https://openalex.org/C26760741","wikidata":"https://www.wikidata.org/wiki/Q160402","display_name":"Perception","level":2,"score":0.42640000581741333},{"id":"https://openalex.org/C2780343955","wikidata":"https://www.wikidata.org/wiki/Q333173","display_name":"Surprise","level":2,"score":0.4185999929904938},{"id":"https://openalex.org/C22927095","wikidata":"https://www.wikidata.org/wiki/Q1784206","display_name":"Stateful firewall","level":3,"score":0.41440001130104065},{"id":"https://openalex.org/C193702766","wikidata":"https://www.wikidata.org/wiki/Q1414548","display_name":"Concurrency","level":2,"score":0.3869999945163727},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.375},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.366100013256073},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.36579999327659607},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.35010001063346863},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.3319999873638153},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.3269999921321869},{"id":"https://openalex.org/C63584917","wikidata":"https://www.wikidata.org/wiki/Q333286","display_name":"Bounding overwatch","level":2,"score":0.3188999891281128},{"id":"https://openalex.org/C2780186347","wikidata":"https://www.wikidata.org/wiki/Q11414","display_name":"Subnetwork","level":2,"score":0.31470000743865967},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.3068999946117401},{"id":"https://openalex.org/C108154423","wikidata":"https://www.wikidata.org/wiki/Q1469792","display_name":"Salience (neuroscience)","level":2,"score":0.30489999055862427},{"id":"https://openalex.org/C169900460","wikidata":"https://www.wikidata.org/wiki/Q2200417","display_name":"Cognition","level":2,"score":0.29350000619888306},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.29249998927116394},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.290800005197525},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2867000102996826},{"id":"https://openalex.org/C130440534","wikidata":"https://www.wikidata.org/wiki/Q14946528","display_name":"Conflation","level":2,"score":0.2786000072956085},{"id":"https://openalex.org/C55282118","wikidata":"https://www.wikidata.org/wiki/Q252683","display_name":"Snapshot (computer storage)","level":2,"score":0.2782999873161316},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.27250000834465027},{"id":"https://openalex.org/C2777548347","wikidata":"https://www.wikidata.org/wiki/Q5456937","display_name":"Flagging","level":2,"score":0.2721000015735626},{"id":"https://openalex.org/C70437156","wikidata":"https://www.wikidata.org/wiki/Q7228652","display_name":"Pooling","level":2,"score":0.27059999108314514},{"id":"https://openalex.org/C108650721","wikidata":"https://www.wikidata.org/wiki/Q1783253","display_name":"Counterfactual thinking","level":2,"score":0.267300009727478},{"id":"https://openalex.org/C162262903","wikidata":"https://www.wikidata.org/wiki/Q343527","display_name":"Allocator","level":2,"score":0.26510000228881836},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.26440000534057617},{"id":"https://openalex.org/C140547941","wikidata":"https://www.wikidata.org/wiki/Q7797194","display_name":"Threat model","level":2,"score":0.26269999146461487},{"id":"https://openalex.org/C2776834041","wikidata":"https://www.wikidata.org/wiki/Q25346349","display_name":"Execution model","level":2,"score":0.26100000739097595},{"id":"https://openalex.org/C177212765","wikidata":"https://www.wikidata.org/wiki/Q627335","display_name":"Workflow","level":2,"score":0.25949999690055847},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.2583000063896179},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.25589999556541443}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.23483","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.23483","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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.2603.23483","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.23483","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Agentic":[0,11],"multimodal":[1],"large":[2,143],"language":[3],"models":[4],"(MLLMs)":[5],"(e.g.,":[6],"OpenAI":[7],"o3":[8],"and":[9,27,42,154],"Gemini":[10],"Vision)":[12],"achieve":[13],"remarkable":[14],"reasoning":[15],"capabilities":[16],"through":[17],"iterative":[18],"visual":[19],"tool":[20,88],"invocation.":[21],"However,":[22],"the":[23,80,110,128,132,137,142,163],"cascaded":[24],"perception,":[25],"reasoning,":[26],"tool-calling":[28],"loops":[29],"introduce":[30,99],"significant":[31],"sequential":[32,61],"overhead.":[33],"This":[34],"overhead,":[35],"termed":[36],"agentic":[37,164],"depth,":[38],"incurs":[39],"prohibitive":[40],"latency":[41],"seriously":[43],"limits":[44],"system-level":[45],"concurrency.":[46],"To":[47,93],"this":[48,60,95],"end,":[49],"we":[50,98,120],"propose":[51],"SpecEyes,":[52],"an":[53],"agentic-level":[54],"speculative":[55,76,96],"acceleration":[56],"framework":[57],"that":[58,67,126,157],"breaks":[59],"bottleneck.":[62],"Our":[63],"key":[64],"insight":[65],"is":[66],"a":[68,75,100,122],"lightweight,":[69],"tool-free":[70],"MLLM":[71],"can":[72],"serve":[73],"as":[74],"planner":[77],"to":[78,135,173],"predict":[79],"execution":[81,140],"trajectory,":[82],"enabling":[83],"early":[84],"termination":[85],"of":[86,131,141],"expensive":[87],"chains":[89],"without":[90,115],"sacrificing":[91],"accuracy.":[92],"regulate":[94],"planning,":[97],"cognitive":[101],"gating":[102],"mechanism":[103],"based":[104],"on":[105,150],"answer":[106],"separability,":[107],"which":[108],"quantifies":[109],"model's":[111],"confidence":[112],"for":[113],"self-verification":[114],"requiring":[116],"oracle":[117],"labels.":[118],"Furthermore,":[119],"design":[121],"heterogeneous":[123],"parallel":[124],"funnel":[125],"exploits":[127],"stateless":[129],"concurrency":[130],"small":[133],"model":[134],"mask":[136],"stateful":[138],"serial":[139],"model,":[144],"maximizing":[145],"system":[146],"throughput.":[147],"Extensive":[148],"experiments":[149],"V*":[151],"Bench,":[152],"HR-Bench,":[153],"POPE":[155],"demonstrate":[156],"SpecEyes":[158],"achieves":[159],"1.1-3.35x":[160],"speedup":[161],"over":[162],"baseline":[165],"while":[166],"preserving":[167],"or":[168],"even":[169],"improving":[170],"accuracy":[171],"(up":[172],"+6.7%),":[174],"thereby":[175],"boosting":[176],"serving":[177],"throughput":[178],"under":[179],"concurrent":[180],"workloads.":[181]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-03-26T00:00:00"}