Learning to Sync at the Edge: Client-Side Adaptive Synchronization for Federated Learning

Ebasa Temesgen; Ammar Ahmed; Nathan Lechamo; Ali Anwar; Maria Gini

cs.DCOct 9, 2026

Learning to Sync at the Edge: Client-Side Adaptive Synchronization for Federated Learning

Authors: Ebasa Temesgen, Ammar Ahmed, Nathan Lechamo, Ali Anwar, Maria Gini

Organizations: Department of Computer Science & Engineering, University of Minnesota, Minneapolis, MN, USA · Department of Computer Science, University of Texas at San Antonio, San Antonio, TX, USA

Abstract

The scalability and performance of edge federated learning are fundamentally constrained by synchronization bottlenecks. Clients operate under heterogeneous compute, time-varying communication conditions, and, in some settings, privacy constraints, yet synchronization is often still handled with fixed schedules. However, synchronization decisions are often made outside the client, even though the client directly observes local progress, resource state, and communication conditions. We present a client-side reinforcement learning synchronization framework in which each client decides during training whether to upload, download, synchronize bidirectionally, or defer communication. The framework places a lightweight scheduler alongside the local model and formulates synchronization as a resource-aware control problem driven by local learning and system state. We study this framework on two representative edge workloads, spatiotemporal traffic forecasting and image classification. Across both settings, adaptive client-side synchronization improves the accuracy-cost trade-off, reducing system cost in traffic forecasting while maintaining competitive cost and improving final accuracy in image classification. We also consider privacy-constrained settings, where synchronization must account for not only latency and energy but also privacy expenditure. Overall, our results show that synchronization in edge federated learning is better treated as an adaptive client-side control problem than only as a server-side coordination mechanism.

Figures & tables

Fig. 1: Server-side versus client-side communication control in federated learning. (a) Prior work typically adapts communication from the server. (b) Our approach places a lightweight scheduler at each client to make synchronization decisions from local state.

Method	Control point	Controller	Decision
FedAsync	Aggregation	Server	Apply updates
FedBuff	Aggregation	Server	Buffer updates
Oort	Participation	Server	Select clients
FedMarl	Participation	Server	Schedule clients
SCAFFOLD	Optimization	Server/client	Correct drift
FedProx	Optimization	Client-objective	Stabilize training

TABLE I: Communication-control mechanisms in federated learning.

Fig. 2: System overview of the proposed client-side synchronization scheduler. Each client trains a local task model on private data and uses an RL policy to decide whether to skip, upload, download, or synchronize bidirectionally based on local learning state and environment conditions. The server maintains the global model and aggregates received client updates.

Fig. 3: Cost accounting at one client’s decision.

Group	Features
Comm.	Availability, channel regime, transfer time and energy
Learning	Loss, loss change, dirty flag, latest compute cost
Sync.	Upload/download age, local/server versions, staleness
Privacy	Spent/remaining budget and most recent increment

TABLE II: Client observation supplied to the shared policy.

Fig. 4: Observation and action-selection pipeline for the client-side RL scheduler.

Fig. 5: Client-size balance for five-way PeMSD7(M) partitioning (228 sensors) [ 30 ] .

Setting	Value
PPO objective	$\gamma=0.99$ , GAE $\lambda=0.95$ , clip $0.2$
PPO optimizer	LR $3\!\times\!10^{-4}$ , 6 epochs, batch 64, entropy $0.006$ , value coefficient $0.5$
Channel	$B=10^{5}$ Hz; multipliers $1.5/1.0/0.45$ with probabilities $0.35/0.45/0.20$
Device	$P_{\rm tx}=1.5$ W, $P_{\rm rx}=0.8$ W; $f_{m}\sim U(0.6,1.0)$ GHz
Privacy	DP; $C=1$ , $\sigma=6$ , $\delta=10^{-5}$ , $\varepsilon_{\max}=12$
Task weights	$\lambda_{\rm loss}=0.35$ , $\lambda_{\rm time}=0.75$ , $\lambda_{\rm ene}=1.0$

TABLE III: Scheduler and environment configuration.

Method	MAE	RMSE	MAPE	Latency (s)	Energy (J)	Uploaded (GB)	Uploads	Downloads	Agg.
DQN	20.44	30.78	14.85	$1.4\!\times\!10^{5}$	$1.8\!\times\!10^{5}$	28.37	6,439	3,552	6,226
FedAsync	20.51	31.34	15.33	$1.4\!\times\!10^{5}$	$1.9\!\times\!10^{5}$	30.56	6,936	3,657	6,712
FedBuff	22.12	32.67	18.40	$1.2\!\times\!10^{5}$	$1.7\!\times\!10^{5}$	31.89	7,237	749	701
Oort	23.00	34.40	22.88	$1.4\!\times\!10^{5}$	$1.9\!\times\!10^{5}$	29.47	6,689	3,873	6,456
PPO	19.64	29.92	16.50	$6.8\!\times\!10^{4}$	$8.1\!\times\!10^{4}$	10.59	2,404	2,506	2,327

TABLE IV: Traffic forecasting on PeMS-08 with four clients over 200 episodes. The validation-selected run is averaged over the evaluation horizon; lower is better for errors and costs.

Method	Accuracy	Latency/episode	Energy/episode	Uploaded/episode	Uploads	Downloads
FedAsync	81.00%	401.0 s	594.4 J	0.111 GB	39,602	1,227
FedBuff	80.81%	394.3 s	587.0 J	0.111 GB	39,666	768
Oort	81.10%	399.2 s	590.9 J	0.111 GB	39,572	1,334
DQN	62.09%	436.3 s	599.2 J	0.103 GB	36,504	9,835
Always-sync	70.66%	731.1 s	857.6 J	0.111 GB	39,610	39,319
PPO	84.48%	406.3 s	595.7 J	0.111 GB	39,464	2,273

TABLE V: Image classification: 20 clients, 300 episodes; means across runs. Lower is better except accuracy.

Fig. 6: Image-workload test-accuracy. PPO reaches the highest final accuracy; DQN converges more slowly under the same 256-decision rollout horizon.

Fig. 7: FedSTAE forecasting quality against representative federated baselines. Short and long horizons correspond to 5 and 60 minutes.

	FedGRU			FedSTAE
Clients	MAE	RMSE	MAPE	MAE	RMSE	MAPE
2	58.58	78.26	61.54	18.78	29.21	16.06
4	66.39	86.94	82.34	19.10	29.71	15.79
6	61.39	82.94	75.23	19.12	29.42	16.49
8	60.01	79.96	83.41	19.93	30.60	17.28
16	56.86	75.31	76.01	20.30	30.50	18.42

TABLE VI: Long-horizon PeMS-08 forecasting as client count increases.

Variant	Acc.	Latency	Energy	Uploaded
Full	85.31%	400.1 s	588.9 J	0.111 GB
w/o bidirectional	75.31%	356.0 s	528.7 J	0.100 GB
w/o bidirectional + learning state	73.75%	351.8 s	521.6 J	0.099 GB
w/o bidirectional + privacy state	74.06%	356.8 s	529.7 J	0.100 GB
w/o bidirectional + sync state	72.81%	355.3 s	527.0 J	0.100 GB

TABLE VII: Single-run image-policy ablations after 300 episodes.

Fig. 8: PPO variant trade-off across 300-episode runs. Accuracy is reevaluated on the full test set using each run’s validation-selected checkpoint; latency covers the full run. All five variants spend the same privacy budget (

\varepsilon=8.06

) , so privacy is not encoded. Shading indicates lower latency and higher accuracy as visual guides.

Federated Learning

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

See all 951 papers

Robust Federated Learning Under Real-World Client Churn
Jul 8, 2026Dhruv Garg, Neha Lakhani, Debopam Sanyal +3Federated Learning Aggregation Communication-Efficient Distributed Training
Georgia Tech USA · Cisco Research USA
Collate: Collaborative Neural Network Learning for Latency-Critical Edge Systems
Jul 9, 2026Shuo Huai, Di Liu, Hao Kong +5Efficient Neural Network Inference Federated Learning Aggregation
School of Computer Science and Engineering, Nanyang Technological University, Singapore · HP-NTU Digital Manufacturing Corporate Lab, Nanyang Technological University, Singapore · HP Inc., Palo Alto, California, USA
ASFL: An Adaptive Model Splitting and Resource Allocation Framework for Split Federated Learning
Feb 19, 2026Chuiyang Meng, Ming Tang, Vincent W. S. WongSplit Federated Learning Energy-Efficient ML
Department of Electrical and Computer Engineering, The University of British Columbia, Vancouver, Canada · Department of Computer Science and Engineering, Southern University of Science and Technology, China

Jul 8, 2026J/K move · Enter open · S save

Robust Federated Learning Under Real-World Client Churn

Dhruv Garg, Neha Lakhani, Debopam Sanyal, Myungjin Lee+2

Georgia Tech USA · Cisco Research USA

Federated Learning Aggregation Communication-Efficient Distributed Training Client Selection in Federated Learning Asynchronous Federated Learning Federated Learning

Open paper AskPDF arXiv

Federated Learning (FL) enables training shared models on private, on-device data, but production deployments remain constrained to slow, multi-day refresh cycles due to the complexity of coordinating massive client populations. For applications such as feed ranking, ad targeting, and personalized recommendation, model freshness: the ability to rapidly adapt to new user-local data is critical for maximizing objectives like click-through rate. This lag leaves models stale and unresponsive to volatile data distributions driven by viral trends and shifting user intent. Bridging this gap requires addressing three challenges overlooked by existing FL systems: transient client availability, dynamic data heterogeneity, and delays between model predictions and observable outcomes. We present FeLiX, an FL orchestration framework that minimizes wall-clock time-to-target accuracy on live interaction streams. FeLiX introduces three primitives: (i) streaming-aware availability tiers that leverage lightweight telemetry to identify ready clients at scale; (ii) fresh-utility selection, a dual-tier mechanism that prioritizes statistically valuable updates from devices able to meet tight refresh deadlines; and (iii) informativeness-aware, delay-robust aggregation that incorporates late, high-value updates containing ground-truth outcomes without biasing the global model toward stale distributions. Unlike prior systems that rely on unrealistic oracular knowledge of client availability, FeLiX achieves near-oracular performance in real-world settings. Across CIFAR-10, Google Speech, and realistic low-availability traces, FeLiX reduces wall-clock time-to-target accuracy by up to 2.37X while reducing communication bandwidth by 1.30X compared to state-of-the-art synchronous and asynchronous FL baselines.

Abstract

Federated Learning

Explore similar work

Robust Federated Learning Under Real-World Client Churn

Collate: Collaborative Neural Network Learning for Latency-Critical Edge Systems

ASFL: An Adaptive Model Splitting and Resource Allocation Framework for Split Federated Learning