FeLiX is presented, an FL orchestration framework that minimizes wall-clock time-to-target accuracy on live interaction streams and achieves near-oracular performance in real-world settings.
Abstract
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.
Federated learning (FL) on heterogeneous edge networks faces a fundamental tension: standard aggregation protocols assume client homogeneity, yet real-world edge deployments span device tiers with $7 \times$ compute and $\mathbf{1 0} \times$ bandwidth disparities. Slow clients become stragglers that stall synchronous rounds, while uniform gradient compression degrades accuracy on bandwidth-constrained devices. This paper presents FedEdge-Adapt, a novel adaptive federated learning framework that jointly addresses device heterogeneity, straggler mitigation, and communication efficiency without sacrificing model quality. FedEdge-Adapt introduces three tightly coupled mechanisms: (1) tier-aware gradient compression that applies device-class-specific sparsification ratios, (2) drift-corrected aggregation that reweights client updates based on staleness and data heterogeneity, and (3) predictive client selection that anticipates dropout-prone devices using a lightweight resource oracle. We evaluate FedEdge-Adapt on a 30-node heterogeneous edge network over 150 communication rounds using the CIFAR-10 dataset under non-IID distributions ($\alpha=0.5$ Dirichlet) and compare against FedAvg, FedProx, and SCAFFOLD baselines. FedEdge-Adapt achieves 85.44% global accuracy, a 6.85 percentage-point improvement over FedAvg, while simultaneously reducing round latency by $\mathbf{6 7. 4 \%}$, communication overhead by $\mathbf{3 4. 0 \%}$, and client dropout rate by $\mathbf{5 4. 0 \%}$. Convergence is reached in 18 rounds versus 31 for FedAvg. Extended experiments across 100+ rounds confirm long-term stability with no late-stage divergence.
Saher Elsayed, Mohamed Ali, Samer Abubaker et al.· Annual International Compute...· 0 citations
This work proposes a data-Quality-aware aggregation framework by introducing an Evolutionary-computation-inspired de-sign into Federated learning ( FedEvoQ), with a lightweight dual-branch architecture.
Federated Learning (FL) is increasingly adopted as an alternative to centralized Machine Learning (ML) techniques, as it allows clients to preserve the privacy of their data. However, FL systems pose new challenges in terms of adaptation, as design choices are conditioned by client characteristics and network conditions, thus necessitating adaptive strategies that elaborate on such a different operational environment. Previous work introduces a set of architectural patterns to support practitioners at design time, but their effectiveness has only been investigated when statically activated throughout the FL process. This work presents a novel FL framework, namely FLiP , where a subset of the aforementioned patterns are dynamically and adaptively toggled in response to evolving performance metrics and boundary conditions. We empirically evaluate FLiP across multiple federation configurations and two learning tasks, considering both static and dynamic conditions. Results indicate that dynamically toggling architectural patterns can be beneficial under specific conditions, with cases leading to an improvement of up to \(10\%\) in learning accuracy, at the cost of negligible overhead at deployment time.
L. Baresi, Ivan Compagnucci, Livia Lestingi et al.· SEAMS@ICSE· 1 citation
A novel FL framework is presented, FedPhoenix, that stochastically re-sets partial parameters in each round to destroy some features of the global model, guiding FL training to learn multiple generalized features for inference rather than specific overfitting features.
Jiahao Wu, Ming Hu, Yanxin Yang et al.· Advances in Neural Informati...· 1 citation