Communication-Efficient Federated Domain Generalization in Edge Networks via Dynamic Prototype Regulation

Yuankun Xia; Hui Wang; Ling Yuan; Ziyu Hong; Di Yin; Yufeng Zhou

cs.LGOct 19, 2026

Communication-Efficient Federated Domain Generalization in Edge Networks via Dynamic Prototype Regulation

Authors: Yuankun Xia, Hui Wang, Ling Yuan, Ziyu Hong, Di Yin, Yufeng Zhou

Organizations: School of Computer Science and Technology, Zhejiang Normal University, Jinhua 321004, China

Abstract

Federated learning (FL) is a distributed paradigm for Internet of Things (IoT) edge networks. However, deploying Vision-Language Models (VLMs) over wireless edge networks presents distinct challenges: large model parameters increase uplink transmission latency over bandwidth-constrained fading channels, and non-independent and identically distributed (Non-IID) data exacerbates client drift. Existing feature alignment solutions often introduce auxiliary computations that consume time slots allocated for wireless parameter transmission, increasing the likelihood of straggler dropouts under system delay constraints. To address these issues, we propose DPGR-GCAM, a communication-efficient federated domain generalization framework. By employing a dual-parameter-efficient fine-tuning (PEFT) architecture, we design a channel-aware tiered transmission mechanism that adaptively truncates payloads to maximize node survival rates under fading conditions. To mitigate client drift, we introduce the Gradient-Conflict Aware Modulation (GCAM) mechanism. GCAM operates within the low-dimensional feature manifold to dynamically resolve multi-objective optimization conflicts with minimal edge computation latency, thereby reducing computation-induced network outages. Experiments demonstrate that DPGR-GCAM improves cross-domain generalization and sustains convergence under network fading scenarios with a 10% client participation rate.

Figures & tables

Fig. 1: The overall workflow of the proposed DPGR-GCAM framework for federated domain generalization over wireless edge networks. The system comprises three stages: (1) Global broadcast of lightweight tunable parameters and class prototypes; (2) Local training, where the GCAM mechanism dynamically resolves feature-level optimization conflicts to regulate the visual adapter update; and (3) Channel-aware aggregation, which adaptively truncates the uplink payload based on real-time Shannon capacity to maximize node survival rates under fading channels.

$\alpha$	Methods	DomainNet							Office-Caltech10
$\alpha$	Methods	c	i	p	q	r	s	avg	a	w	d	c	avg
0.01	FedAvg [ 17 ]	81.37	59.28	77.32	26.69	89.29	75.53	68.25	93.75	89.66	93.55	93.78	92.69
	FedProto [ 18 ]	83.68	65.54	76.79	26.86	88.38	76.50	69.63	94.79	93.10	93.55	95.33	94.19
	FedCLIP [ 9 ]	82.99	58.60	76.53	25.83	89.26	75.16	68.06	95.83	93.10	93.55	93.33	93.95
	PromptFL [ 8 ]	85.94	38.36	62.79	21.71	79.96	67.39	59.36	96.35	97.56	98.98	94.22	96.78
	FedAPT [ 19 ]	82.69	39.69	59.31	21.60	77.34	65.59	57.70	97.91	98.28	100.00	96.44	98.16

TABLE I: Cross-Domain Generalization Accuracy (%) on DomainNet and Office-Caltech10 under Varying Data Heterogeneity (

\alpha

)

Panel of Figure 3 — Fig. 2: Robustness evaluation under intermittent connectivity on different datasets. (a) Moderate-scale Office-Caltech10. (b) Large-scale DomainNet.

Method	Trainable	Comm. Payload	Peak Memory
Method	Params (M)	(MB)	(MB)
Full Fine-tuning	127.68	487.04	10,968
PromptFL [ 8 ]	2.83	10.78	6,124
FedAPT [ 19 ]	3.09	11.78	6,127
FedOTP [ 20 ]	5.65	10.78	6,147
Ours	3.35	12.78	6,125

TABLE II: Communication and Computation Overhead on Edge Nodes

Components	DomainNet	Office-Caltech10
PEFT only	71.45	95.23
PEFT + $\mathcal{L}_{reg}$	72.84	96.43
w/o Adapter	74.16	97.21
DPGR-GCAM (Full)	76.04	98.93

TABLE III: Effectiveness of Key Components in DPGR-GCAM

Federated Learning

Federated learning trains shared models across data holders without requiring them to pool their raw data.

See all 921 papers

EdgeFLow: Serverless Federated Learning via Sequential Model Migration in Edge Networks
Mar 3, 2026Yuchen Shi, Qijun Hou, Pingyi Fan +1Federated Learning Communication-Efficient Distributed Training
Dept. of Electronic Engineering, BNRist, Tsinghua University. · Dept. of Electronic and Computer Engineering, HKUST.
PFAdapter: Hierarchical LoRA Decomposition for Personalized Federated MLLMs
Jul 13, 2026Jing Liu, Kun Yang, Yan Wang +5Multimodal Federated Learning Federated Learning Aggregation
College of Future Information Technology, Fudan University, Shanghai 200433, China · Division of Natural and Applied Sciences, Duke Kunshan University, Suzhou 215316, China · Department of Electrical and Computer Engineering, The University of British Columbia, BC V6T 1Z4, Canada +8
FedZMG: Efficient Client-Side Optimization in Federated Learning
Feb 20, 2026Fotios Zantalis, Evangelos Zervas, Grigorios KoulourasFederated Learning Non-IID Federated Learning
TelSiP Research Laboratory, Department of Electrical and Electronic Engineering, School of Engineering, University of West Attica, Ancient Olive Grove Campus, 250 Thivon Str., GR-12241 Athens, Greece

Mar 3, 2026J/K move · Enter open · S save

EdgeFLow: Serverless Federated Learning via Sequential Model Migration in Edge Networks

Yuchen Shi, Qijun Hou, Pingyi Fan, Khaled B. Letaief

Dept. of Electronic Engineering, BNRist, Tsinghua University. · Dept. of Electronic and Computer Engineering, HKUST.

Federated Learning Communication-Efficient Distributed Training

Open paper AskPDF arXiv

Federated Learning (FL) has emerged as a transformative distributed learning paradigm in the era of Internet of Things (IoT), reconceptualizing data processing methodologies. However, FL systems face significant communication bottlenecks due to inevitable client-server data exchanges and long-distance transmissions. This work presents EdgeFLow, an innovative FL framework that redesigns the system topology by replacing traditional cloud servers with sequential model migration between edge base stations. By conducting model aggregation and propagation exclusively at edge clusters, EdgeFLow eliminates cloud-based transmissions and substantially reduces global communication overhead. We provide rigorous convergence analysis for EdgeFLow under non-convex objectives and non-IID data distributions, extending classical FL convergence theory. Experimental results across various configurations validate the theoretical analysis, demonstrating that EdgeFLow achieves comparable accuracy improvements while significantly reducing communication costs. As a systemic architectural innovation for communication-efficient FL, EdgeFLow establishes a foundational framework for future developments in IoT and edge-network learning systems.

Abstract

Federated Learning

Explore similar work

EdgeFLow: Serverless Federated Learning via Sequential Model Migration in Edge Networks

PFAdapter: Hierarchical LoRA Decomposition for Personalized Federated MLLMs

FedZMG: Efficient Client-Side Optimization in Federated Learning