{"id":"https://openalex.org/W7162536842","doi":"https://doi.org/10.48550/arxiv.2605.26647","title":"More Expressive Feedforward Layers: Part I. Token-Adaptive Mixing of Activations","display_name":"More Expressive Feedforward Layers: Part I. Token-Adaptive Mixing of Activations","publication_year":2026,"publication_date":"2026-05-26","ids":{"openalex":"https://openalex.org/W7162536842","doi":"https://doi.org/10.48550/arxiv.2605.26647"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.26647","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.26647","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.2605.26647","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137159433","display_name":"Mingze Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Mingze","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137104498","display_name":"Jinbo Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Jinbo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033298018","display_name":"Yikuan Xia","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xia, Yikuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137100539","display_name":"Kai Shen","orcid":"https://orcid.org/0009-0003-2127-9207"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shen, Kai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5012790743","display_name":"Shu Zhong","orcid":"https://orcid.org/0000-0002-1820-6424"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhong, Shu","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/T10028","display_name":"Topic Modeling","score":0.478300005197525,"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/T10028","display_name":"Topic Modeling","score":0.478300005197525,"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/T11948","display_name":"Machine Learning in Materials Science","score":0.09600000083446503,"subfield":{"id":"https://openalex.org/subfields/2505","display_name":"Materials Chemistry"},"field":{"id":"https://openalex.org/fields/25","display_name":"Materials Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.06920000165700912,"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/expressivity","display_name":"Expressivity","score":0.6922000050544739},{"id":"https://openalex.org/keywords/feed-forward","display_name":"Feed forward","score":0.6632999777793884},{"id":"https://openalex.org/keywords/nonlinear-system","display_name":"Nonlinear system","score":0.59170001745224},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.5784000158309937},{"id":"https://openalex.org/keywords/mixing","display_name":"Mixing (physics)","score":0.5164999961853027},{"id":"https://openalex.org/keywords/function","display_name":"Function (biology)","score":0.3953999876976013},{"id":"https://openalex.org/keywords/ranging","display_name":"Ranging","score":0.34299999475479126}],"concepts":[{"id":"https://openalex.org/C92811239","wikidata":"https://www.wikidata.org/wiki/Q20998670","display_name":"Expressivity","level":2,"score":0.6922000050544739},{"id":"https://openalex.org/C38858127","wikidata":"https://www.wikidata.org/wiki/Q5441228","display_name":"Feed forward","level":2,"score":0.6632999777793884},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.630299985408783},{"id":"https://openalex.org/C158622935","wikidata":"https://www.wikidata.org/wiki/Q660848","display_name":"Nonlinear system","level":2,"score":0.59170001745224},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5784000158309937},{"id":"https://openalex.org/C138777275","wikidata":"https://www.wikidata.org/wiki/Q6884054","display_name":"Mixing (physics)","level":2,"score":0.5164999961853027},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.41839998960494995},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.40639999508857727},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.3953999876976013},{"id":"https://openalex.org/C115051666","wikidata":"https://www.wikidata.org/wiki/Q6522493","display_name":"Ranging","level":2,"score":0.34299999475479126},{"id":"https://openalex.org/C38365724","wikidata":"https://www.wikidata.org/wiki/Q4677469","display_name":"Activation function","level":3,"score":0.34060001373291016},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.3296000063419342},{"id":"https://openalex.org/C47702885","wikidata":"https://www.wikidata.org/wiki/Q5441227","display_name":"Feedforward neural network","level":3,"score":0.3208000063896179},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.30399999022483826},{"id":"https://openalex.org/C2779664074","wikidata":"https://www.wikidata.org/wiki/Q3518405","display_name":"Terminal (telecommunication)","level":2,"score":0.28110000491142273},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.27810001373291016},{"id":"https://openalex.org/C149629883","wikidata":"https://www.wikidata.org/wiki/Q660926","display_name":"Fraction (chemistry)","level":2,"score":0.26019999384880066},{"id":"https://openalex.org/C204241405","wikidata":"https://www.wikidata.org/wiki/Q461499","display_name":"Transformation (genetics)","level":3,"score":0.258899986743927}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.26647","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.26647","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.2605.26647","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.26647","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":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Feedforward":[0],"network":[1],"(FFN)":[2],"layers":[3],"account":[4],"for":[5,98,199],"a":[6,38,60,66,194],"large":[7,16],"fraction":[8],"of":[9,57,68,95],"parameters":[10,155],"and":[11,25,101,115,146,161,171,183,196],"nonlinear":[12,46,134],"expressivity":[13,130,202],"in":[14,203],"Transformer-based":[15],"language":[17,148],"models":[18,149],"(LLMs).":[19],"Despite":[20],"the":[21,44,77,128],"evolution":[22],"from":[23,132,151],"ReLU":[24],"GELU":[26],"to":[27,48,153],"gated":[28],"variants":[29],"such":[30],"as":[31],"SwiGLU,":[32],"most":[33],"FFN":[34,62,201],"designs":[35],"still":[36],"use":[37],"single":[39],"fixed":[40],"activation":[41,69,96,191],"function,":[42],"applying":[43],"same":[45,78],"transformation":[47],"all":[49],"tokens.":[50],"In":[51],"this":[52],"work,":[53],"we":[54,85,105,137],"propose":[55],"Mixture":[56],"Activations":[58],"(MoA),":[59],"token-adaptive":[61,190],"design":[63],"that":[64,189],"mixes":[65],"dictionary":[67],"functions":[70,97],"using":[71],"lightweight":[72],"input-dependent":[73,133],"gates":[74],"while":[75,122],"sharing":[76],"linear":[79,93],"projections.":[80],"As":[81],"an":[82],"input-independent":[83],"counterpart,":[84],"also":[86],"introduce":[87],"learnable":[88],"activations":[89],"(LA),":[90],"which":[91],"form":[92],"combinations":[94],"both":[99],"ReLU-type":[100],"SwiGLU-type":[102],"FFNs.":[103],"Theoretically,":[104],"establish":[106],"strict":[107],"finite-width":[108],"expressive":[109],"separations":[110],"among":[111],"fixed-activation":[112,120],"FFNs,":[113,121],"LA,":[114,126],"MoA:":[116],"LA":[117],"strictly":[118,124],"contains":[119,125],"MoA":[123,139,165],"with":[127,180],"additional":[129],"arising":[131],"hybridization.":[135],"Empirically,":[136],"evaluate":[138],"through":[140],"extensive":[141],"pre-training":[142],"experiments":[143],"on":[144],"dense":[145],"MoE":[147],"ranging":[150],"0.12B":[152],"2B":[154],"under":[156],"different":[157],"token":[158],"budgets,":[159],"optimizers,":[160],"learning":[162],"rate":[163],"schedules.":[164],"consistently":[166],"achieves":[167],"lower":[168],"terminal":[169],"loss":[170],"exhibits":[172],"more":[173],"favorable":[174],"scaling":[175],"behavior":[176],"than":[177],"well-tuned":[178],"baselines,":[179],"minimal":[181],"parameter":[182],"computational":[184],"overhead.":[185],"These":[186],"results":[187],"suggest":[188],"mixing":[192],"is":[193],"simple":[195],"effective":[197],"mechanism":[198],"improving":[200],"LLMs.":[204]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-28T00:00:00"}