ADF-LoRA: Alternating Low-Rank Aggregation for Decentralized Federated Fine-Tuning Article Swipe

View

Wang Xiaoyu , Li Xiaotian , Zhou Zhixiang , Li Chen , Liu, Yong ·

YOU? · · 2025 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2511.18291

This paper revisits alternating low-rank updates for federated fine-tuning and examines their behavior in decentralized federated learning (DFL). While alternating the LoRA matrices has been shown to stabilize aggregation in centralized FL, extending this mechanism to decentralized, peer-to-peer communication introduces new challenges due to phase-state mismatch and block-wise divergence across clients. We introduce ADF-LoRA, which synchronizes the update of only one low-rank matrix per round and mixes both matrices to maintain more consistent parameter states under decentralized propagation. This design preserves the cross-term suppression effect of alternating updates while improving stability in serverless topologies. We provide a convergence analysis under standard smoothness assumptions and evaluate ADF-LoRA on multiple GLUE tasks. Experiments show that ADF-LoRA achieves faster and smoother convergence and delivers the highest average accuracy across tasks, outperforming existing LoRA variants in decentralized FL by a consistent margin.

Related Topics

Computer Science

Smoothness

Federated Learning

Theoretical Computer Science

Concepts

Computer science Convergence (economics) Stability (learning theory) Divergence (linguistics) Smoothness Distributed computing Decentralised system Matrix (chemical analysis) Mechanism (biology) Mathematical optimization Federated learning State (computer science) Sequence (biology) Theoretical computer science Scheme (mathematics)

Metadata

Type: preprint
Landing Page: https://doi.org/10.48550/arxiv.2511.18291
OA Status: green
OpenAlex ID: https://openalex.org/W7106647550

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W7106647550

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48550/arxiv.2511.18291

Digital Object Identifier
Title: ADF-LoRA: Alternating Low-Rank Aggregation for Decentralized Federated Fine-Tuning

Work title
Type: preprint

OpenAlex work type
Publication year: 2025

Year of publication
Publication date: 2025-11-23

Full publication date if available
Authors: Wang Xiaoyu, Li Xiaotian, Zhou Zhixiang, Li Chen, Liu, Yong

List of authors in order
Landing page: https://doi.org/10.48550/arxiv.2511.18291

Publisher landing page
Open access: Yes

Whether a free full text is available
OA status: green

Open access status per OpenAlex
OA URL: https://doi.org/10.48550/arxiv.2511.18291

Direct OA link when available
Concepts: Computer science, Convergence (economics), Stability (learning theory), Divergence (linguistics), Smoothness, Distributed computing, Decentralised system, Matrix (chemical analysis), Mechanism (biology), Mathematical optimization, Federated learning, State (computer science), Sequence (biology), Theoretical computer science, Scheme (mathematics)

Top concepts (fields/topics) attached by OpenAlex
Cited by: 0

Total citation count in OpenAlex

Full payload

id	https://openalex.org/W7106647550
doi	https://doi.org/10.48550/arxiv.2511.18291
ids.doi	https://doi.org/10.48550/arxiv.2511.18291
ids.openalex	https://openalex.org/W7106647550
fwci
type	preprint
title	ADF-LoRA: Alternating Low-Rank Aggregation for Decentralized Federated Fine-Tuning
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
is_xpac	False
apc_list
apc_paid
concepts[0].id	https://openalex.org/C41008148
concepts[0].level	0
concepts[0].score	0.7617626190185547
concepts[0].wikidata	https://www.wikidata.org/wiki/Q21198
concepts[0].display_name	Computer science
concepts[1].id	https://openalex.org/C2777303404
concepts[1].level	2
concepts[1].score	0.7296432256698608
concepts[1].wikidata	https://www.wikidata.org/wiki/Q759757
concepts[1].display_name	Convergence (economics)
concepts[2].id	https://openalex.org/C112972136
concepts[2].level	2
concepts[2].score	0.5864784121513367
concepts[2].wikidata	https://www.wikidata.org/wiki/Q7595718
concepts[2].display_name	Stability (learning theory)
concepts[3].id	https://openalex.org/C207390915
concepts[3].level	2
concepts[3].score	0.5207406878471375
concepts[3].wikidata	https://www.wikidata.org/wiki/Q1230525
concepts[3].display_name	Divergence (linguistics)
concepts[4].id	https://openalex.org/C102634674
concepts[4].level	2
concepts[4].score	0.5159288644790649
concepts[4].wikidata	https://www.wikidata.org/wiki/Q868473
concepts[4].display_name	Smoothness
concepts[5].id	https://openalex.org/C120314980
concepts[5].level	1
concepts[5].score	0.515843391418457
concepts[5].wikidata	https://www.wikidata.org/wiki/Q180634
concepts[5].display_name	Distributed computing
concepts[6].id	https://openalex.org/C205875254
concepts[6].level	3
concepts[6].score	0.4048011004924774
concepts[6].wikidata	https://www.wikidata.org/wiki/Q17156857
concepts[6].display_name	Decentralised system
concepts[7].id	https://openalex.org/C106487976
concepts[7].level	2
concepts[7].score	0.3875787556171417
concepts[7].wikidata	https://www.wikidata.org/wiki/Q685816
concepts[7].display_name	Matrix (chemical analysis)
concepts[8].id	https://openalex.org/C89611455
concepts[8].level	2
concepts[8].score	0.34745675325393677
concepts[8].wikidata	https://www.wikidata.org/wiki/Q6804646
concepts[8].display_name	Mechanism (biology)
concepts[9].id	https://openalex.org/C126255220
concepts[9].level	1
concepts[9].score	0.3005238473415375
concepts[9].wikidata	https://www.wikidata.org/wiki/Q141495
concepts[9].display_name	Mathematical optimization
concepts[10].id	https://openalex.org/C2992525071
concepts[10].level	2
concepts[10].score	0.30017101764678955
concepts[10].wikidata	https://www.wikidata.org/wiki/Q50818671
concepts[10].display_name	Federated learning
concepts[11].id	https://openalex.org/C48103436
concepts[11].level	2
concepts[11].score	0.2937595248222351
concepts[11].wikidata	https://www.wikidata.org/wiki/Q599031
concepts[11].display_name	State (computer science)
concepts[12].id	https://openalex.org/C2778112365
concepts[12].level	2
concepts[12].score	0.2858316898345947
concepts[12].wikidata	https://www.wikidata.org/wiki/Q3511065
concepts[12].display_name	Sequence (biology)
concepts[13].id	https://openalex.org/C80444323
concepts[13].level	1
concepts[13].score	0.2704944908618927
concepts[13].wikidata	https://www.wikidata.org/wiki/Q2878974
concepts[13].display_name	Theoretical computer science
concepts[14].id	https://openalex.org/C77618280
concepts[14].level	2
concepts[14].score	0.2644764482975006
concepts[14].wikidata	https://www.wikidata.org/wiki/Q1155772
concepts[14].display_name	Scheme (mathematics)
keywords[0].id	https://openalex.org/keywords/convergence
keywords[0].score	0.7296432256698608
keywords[0].display_name	Convergence (economics)
keywords[1].id	https://openalex.org/keywords/stability
keywords[1].score	0.5864784121513367
keywords[1].display_name	Stability (learning theory)
keywords[2].id	https://openalex.org/keywords/divergence
keywords[2].score	0.5207406878471375
keywords[2].display_name	Divergence (linguistics)
keywords[3].id	https://openalex.org/keywords/smoothness
keywords[3].score	0.5159288644790649
keywords[3].display_name	Smoothness
keywords[4].id	https://openalex.org/keywords/decentralised-system
keywords[4].score	0.4048011004924774
keywords[4].display_name	Decentralised system
keywords[5].id	https://openalex.org/keywords/matrix
keywords[5].score	0.3875787556171417
keywords[5].display_name	Matrix (chemical analysis)
language
locations[0].id	doi:10.48550/arxiv.2511.18291
locations[0].is_oa	True
locations[0].source.id	https://openalex.org/S4306400194
locations[0].source.issn
locations[0].source.type	repository
locations[0].source.is_oa	True
locations[0].source.issn_l
locations[0].source.is_core	False
locations[0].source.is_in_doaj	False
locations[0].source.display_name	arXiv (Cornell University)
locations[0].source.host_organization	https://openalex.org/I205783295
locations[0].source.host_organization_name	Cornell University
locations[0].source.host_organization_lineage	https://openalex.org/I205783295
locations[0].license
locations[0].pdf_url
locations[0].version
locations[0].raw_type	article
locations[0].license_id
locations[0].is_accepted	False
locations[0].is_published
locations[0].raw_source_name
locations[0].landing_page_url	https://doi.org/10.48550/arxiv.2511.18291
indexed_in	datacite
authorships[0].author.id	https://openalex.org/A1778731936
authorships[0].author.orcid	https://orcid.org/0009-0009-2908-5275
authorships[0].author.display_name	Wang Xiaoyu
authorships[0].author_position	first
authorships[0].raw_author_name	Wang, Xiaoyu
authorships[0].is_corresponding	True
authorships[1].author.id	https://openalex.org/A2231158419
authorships[1].author.orcid
authorships[1].author.display_name	Li Xiaotian
authorships[1].author_position	middle
authorships[1].raw_author_name	Li, Xiaotian
authorships[1].is_corresponding	False
authorships[2].author.id	https://openalex.org/A2025250459
authorships[2].author.orcid
authorships[2].author.display_name	Zhou Zhixiang
authorships[2].author_position	middle
authorships[2].raw_author_name	Zhou, Zhixiang
authorships[2].is_corresponding	False
authorships[3].author.id	https://openalex.org/A1971984624
authorships[3].author.orcid	https://orcid.org/0000-0001-5187-0928
authorships[3].author.display_name	Li Chen
authorships[3].author_position	middle
authorships[3].raw_author_name	Li, Chen
authorships[3].is_corresponding	False
authorships[4].author.id	https://openalex.org/A2058749718
authorships[4].author.orcid	https://orcid.org/0000-0002-0228-8971
authorships[4].author.display_name	Liu, Yong
authorships[4].author_position	last
authorships[4].raw_author_name	Liu, Yong
authorships[4].is_corresponding	False
has_content.pdf	False
has_content.grobid_xml	False
is_paratext	False
open_access.is_oa	True
open_access.oa_url	https://doi.org/10.48550/arxiv.2511.18291
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2025-11-27T00:00:00
display_name	ADF-LoRA: Alternating Low-Rank Aggregation for Decentralized Federated Fine-Tuning
has_fulltext	False
is_retracted	False
updated_date	2025-11-27T01:16:37.896743
primary_topic
cited_by_count	0
locations_count	1
best_oa_location.id	doi:10.48550/arxiv.2511.18291
best_oa_location.is_oa	True
best_oa_location.source.id	https://openalex.org/S4306400194
best_oa_location.source.issn
best_oa_location.source.type	repository
best_oa_location.source.is_oa	True
best_oa_location.source.issn_l
best_oa_location.source.is_core	False
best_oa_location.source.is_in_doaj	False
best_oa_location.source.display_name	arXiv (Cornell University)
best_oa_location.source.host_organization	https://openalex.org/I205783295
best_oa_location.source.host_organization_name	Cornell University
best_oa_location.source.host_organization_lineage	https://openalex.org/I205783295
best_oa_location.license
best_oa_location.pdf_url
best_oa_location.version
best_oa_location.raw_type	article
best_oa_location.license_id
best_oa_location.is_accepted	False
best_oa_location.is_published	False
best_oa_location.raw_source_name
best_oa_location.landing_page_url	https://doi.org/10.48550/arxiv.2511.18291
primary_location.id	doi:10.48550/arxiv.2511.18291
primary_location.is_oa	True
primary_location.source.id	https://openalex.org/S4306400194
primary_location.source.issn
primary_location.source.type	repository
primary_location.source.is_oa	True
primary_location.source.issn_l
primary_location.source.is_core	False
primary_location.source.is_in_doaj	False
primary_location.source.display_name	arXiv (Cornell University)
primary_location.source.host_organization	https://openalex.org/I205783295
primary_location.source.host_organization_name	Cornell University
primary_location.source.host_organization_lineage	https://openalex.org/I205783295
primary_location.license
primary_location.pdf_url
primary_location.version
primary_location.raw_type	article
primary_location.license_id
primary_location.is_accepted	False
primary_location.is_published	False
primary_location.raw_source_name
primary_location.landing_page_url	https://doi.org/10.48550/arxiv.2511.18291
publication_date	2025-11-23
publication_year	2025
referenced_works_count	0
abstract_inverted_index.a	96, 135
abstract_inverted_index.FL	133
abstract_inverted_index.We	51, 94
abstract_inverted_index.by	134
abstract_inverted_index.in	13, 29, 91, 131
abstract_inverted_index.of	58, 85
abstract_inverted_index.on	106
abstract_inverted_index.to	26, 35, 43, 69
abstract_inverted_index.FL,	31
abstract_inverted_index.and	9, 46, 65, 103, 116, 119
abstract_inverted_index.due	42
abstract_inverted_index.for	6
abstract_inverted_index.has	23
abstract_inverted_index.new	40
abstract_inverted_index.one	60
abstract_inverted_index.per	63
abstract_inverted_index.the	20, 56, 81, 121
abstract_inverted_index.GLUE	108
abstract_inverted_index.LoRA	21, 129
abstract_inverted_index.This	0, 78
abstract_inverted_index.been	24
abstract_inverted_index.both	67
abstract_inverted_index.more	71
abstract_inverted_index.only	59
abstract_inverted_index.show	111
abstract_inverted_index.that	112
abstract_inverted_index.this	33
abstract_inverted_index.While	18
abstract_inverted_index.mixes	66
abstract_inverted_index.paper	1
abstract_inverted_index.round	64
abstract_inverted_index.shown	25
abstract_inverted_index.their	11
abstract_inverted_index.under	75, 99
abstract_inverted_index.which	54
abstract_inverted_index.while	88
abstract_inverted_index.(DFL).	17
abstract_inverted_index.across	49, 125
abstract_inverted_index.design	79
abstract_inverted_index.effect	84
abstract_inverted_index.faster	115
abstract_inverted_index.matrix	62
abstract_inverted_index.states	74
abstract_inverted_index.tasks,	126
abstract_inverted_index.tasks.	109
abstract_inverted_index.update	57
abstract_inverted_index.average	123
abstract_inverted_index.highest	122
abstract_inverted_index.margin.	137
abstract_inverted_index.provide	95
abstract_inverted_index.updates	5, 87
abstract_inverted_index.ADF-LoRA	105, 113
abstract_inverted_index.accuracy	124
abstract_inverted_index.achieves	114
abstract_inverted_index.analysis	98
abstract_inverted_index.behavior	12
abstract_inverted_index.clients.	50
abstract_inverted_index.delivers	120
abstract_inverted_index.evaluate	104
abstract_inverted_index.examines	10
abstract_inverted_index.existing	128
abstract_inverted_index.learning	16
abstract_inverted_index.low-rank	4, 61
abstract_inverted_index.maintain	70
abstract_inverted_index.matrices	22, 68
abstract_inverted_index.mismatch	45
abstract_inverted_index.multiple	107
abstract_inverted_index.revisits	2
abstract_inverted_index.smoother	117
abstract_inverted_index.standard	100
abstract_inverted_index.variants	130
abstract_inverted_index.ADF-LoRA,	53
abstract_inverted_index.extending	32
abstract_inverted_index.federated	7, 15
abstract_inverted_index.improving	89
abstract_inverted_index.introduce	52
abstract_inverted_index.mechanism	34
abstract_inverted_index.parameter	73
abstract_inverted_index.preserves	80
abstract_inverted_index.stability	90
abstract_inverted_index.stabilize	27
abstract_inverted_index.block-wise	47
abstract_inverted_index.challenges	41
abstract_inverted_index.consistent	72, 136
abstract_inverted_index.cross-term	82
abstract_inverted_index.divergence	48
abstract_inverted_index.introduces	39
abstract_inverted_index.serverless	92
abstract_inverted_index.smoothness	101
abstract_inverted_index.Experiments	110
abstract_inverted_index.aggregation	28
abstract_inverted_index.alternating	3, 19, 86
abstract_inverted_index.assumptions	102
abstract_inverted_index.centralized	30
abstract_inverted_index.convergence	97, 118
abstract_inverted_index.fine-tuning	8
abstract_inverted_index.phase-state	44
abstract_inverted_index.suppression	83
abstract_inverted_index.topologies.	93
abstract_inverted_index.peer-to-peer	37
abstract_inverted_index.propagation.	77
abstract_inverted_index.synchronizes	55
abstract_inverted_index.communication	38
abstract_inverted_index.decentralized	14, 76, 132
abstract_inverted_index.outperforming	127
abstract_inverted_index.decentralized,	36
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	5
citation_normalized_percentile