{"id":"https://openalex.org/W7153275262","doi":"https://doi.org/10.48550/arxiv.2604.08335","title":"Dead Weights, Live Signals: Feedforward Graphs of Frozen Language Models","display_name":"Dead Weights, Live Signals: Feedforward Graphs of Frozen Language Models","publication_year":2026,"publication_date":"2026-04-09","ids":{"openalex":"https://openalex.org/W7153275262","doi":"https://doi.org/10.48550/arxiv.2604.08335"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.08335","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.08335","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":null,"license_id":null,"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.2604.08335","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5075543992","display_name":"Marcus Armstrong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Armstrong, Marcus","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5080912794","display_name":"Navid Ayoobi","orcid":"https://orcid.org/0009-0003-6841-3645"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ayoobi, Navid","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5078060919","display_name":"Arjun Mukherjee","orcid":"https://orcid.org/0000-0002-8896-604X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mukherjee, Arjun","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/T11273","display_name":"Advanced Graph Neural Networks","score":0.20579999685287476,"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/T11273","display_name":"Advanced Graph Neural Networks","score":0.20579999685287476,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.18559999763965607,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.15029999613761902,"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/feed-forward","display_name":"Feed forward","score":0.6593000292778015},{"id":"https://openalex.org/keywords/signal-flow-graph","display_name":"Signal-flow graph","score":0.6051999926567078},{"id":"https://openalex.org/keywords/encode","display_name":"ENCODE","score":0.4767000079154968},{"id":"https://openalex.org/keywords/residual","display_name":"Residual","score":0.4611000120639801},{"id":"https://openalex.org/keywords/graph","display_name":"Graph","score":0.4293999969959259},{"id":"https://openalex.org/keywords/feedforward-neural-network","display_name":"Feedforward neural network","score":0.42570000886917114},{"id":"https://openalex.org/keywords/backpropagation","display_name":"Backpropagation","score":0.3668999969959259},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.35589998960494995}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6657000184059143},{"id":"https://openalex.org/C38858127","wikidata":"https://www.wikidata.org/wiki/Q5441228","display_name":"Feed forward","level":2,"score":0.6593000292778015},{"id":"https://openalex.org/C166501922","wikidata":"https://www.wikidata.org/wiki/Q1786523","display_name":"Signal-flow graph","level":2,"score":0.6051999926567078},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.536300003528595},{"id":"https://openalex.org/C66746571","wikidata":"https://www.wikidata.org/wiki/Q1134833","display_name":"ENCODE","level":3,"score":0.4767000079154968},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.4611000120639801},{"id":"https://openalex.org/C132525143","wikidata":"https://www.wikidata.org/wiki/Q141488","display_name":"Graph","level":2,"score":0.4293999969959259},{"id":"https://openalex.org/C47702885","wikidata":"https://www.wikidata.org/wiki/Q5441227","display_name":"Feedforward neural network","level":3,"score":0.42570000886917114},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.3792000114917755},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.37290000915527344},{"id":"https://openalex.org/C155032097","wikidata":"https://www.wikidata.org/wiki/Q798503","display_name":"Backpropagation","level":3,"score":0.3668999969959259},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.35589998960494995},{"id":"https://openalex.org/C62611344","wikidata":"https://www.wikidata.org/wiki/Q1062658","display_name":"Node (physics)","level":2,"score":0.3422999978065491},{"id":"https://openalex.org/C51167844","wikidata":"https://www.wikidata.org/wiki/Q4422623","display_name":"Latent variable","level":2,"score":0.33149999380111694},{"id":"https://openalex.org/C2778572836","wikidata":"https://www.wikidata.org/wiki/Q380933","display_name":"Space (punctuation)","level":2,"score":0.3188999891281128},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.27070000767707825},{"id":"https://openalex.org/C193415008","wikidata":"https://www.wikidata.org/wiki/Q639681","display_name":"Network architecture","level":2,"score":0.2678999900817871},{"id":"https://openalex.org/C146380142","wikidata":"https://www.wikidata.org/wiki/Q1137726","display_name":"Directed graph","level":2,"score":0.2651999890804291},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2630000114440918},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.2558000087738037},{"id":"https://openalex.org/C38349280","wikidata":"https://www.wikidata.org/wiki/Q1434290","display_name":"Flow (mathematics)","level":2,"score":0.2500999867916107}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.08335","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.08335","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.08335","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.08335","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"score":0.5558722615242004,"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"We":[0],"present":[1],"a":[2,19,78,97],"feedforward":[3],"graph":[4],"architecture":[5,112],"in":[6],"which":[7],"heterogeneous":[8],"frozen":[9,69,90,144,157],"large":[10],"language":[11],"models":[12,70,91,146],"serve":[13],"as":[14],"computational":[15],"nodes,":[16],"communicating":[17],"through":[18,62,155],"shared":[20,79],"continuous":[21],"latent":[22,39,80],"space":[23,81],"via":[24,60],"learned":[25,141],"linear":[26],"projections.":[27],"Building":[28],"on":[29,115,118,122,143],"recent":[30],"work":[31],"demonstrating":[32],"geometric":[33],"compatibility":[34],"between":[35],"independently":[36],"trained":[37],"LLM":[38],"spaces~\\cite{armstrong2026thinking},":[40],"we":[41],"extend":[42],"this":[43],"finding":[44],"from":[45],"static":[46],"two-model":[47],"steering":[48],"to":[49,163],"end-to-end":[50],"trainable":[51,105],"multi-node":[52],"graphs,":[53],"where":[54],"projection":[55],"matrices":[56],"are":[57],"optimized":[58],"jointly":[59],"backpropagation":[61],"residual":[63],"stream":[64],"injection":[65],"hooks.":[66],"Three":[67],"small":[68],"(Llama-3.2-1B,":[71],"Qwen2.5-1.5B,":[72],"Gemma-2-2B)":[73],"encode":[74],"the":[75,111,125,167],"input":[76],"into":[77,87],"whose":[82,94],"aggregate":[83],"signal":[84],"is":[85,160],"injected":[86],"two":[88],"larger":[89],"(Phi-3-mini,":[92],"Mistral-7B),":[93],"representations":[95],"feed":[96],"lightweight":[98],"cross-attention":[99],"output":[100,168],"node.":[101],"With":[102],"only":[103],"17.6M":[104],"parameters":[106],"against":[107],"approximately":[108],"12B":[109],"frozen,":[110],"achieves":[113],"87.3\\%":[114],"ARC-Challenge,":[116],"82.8\\%":[117],"OpenBookQA,":[119],"and":[120,133,138,150,166],"67.2\\%":[121],"MMLU,":[123],"outperforming":[124,139],"best":[126],"single":[127,145],"constituent":[128],"model":[129,158],"by":[130,147],"11.4,":[131],"6.2,":[132],"1.2":[134],"percentage":[135],"points":[136],"respectively,":[137],"parameter-matched":[140],"classifiers":[142],"9.1,":[148],"5.2,":[149],"6.7":[151],"points.":[152],"Gradient":[153],"flow":[154],"multiple":[156],"boundaries":[159],"empirically":[161],"verified":[162],"be":[164],"tractable,":[165],"node":[169],"develops":[170],"selective":[171],"routing":[172],"behavior":[173],"across":[174],"layer-2":[175],"nodes":[176],"without":[177],"explicit":[178],"supervision.":[179]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-04-11T00:00:00"}