Life sciences · Preprint

FedLBW: A Loss-Based Weighting Strategy for Federated Learning on Non-IID Data in Wireless Networks

arXiv · August 7, 2026

Preliminary Why this matters to you Expert discussion

Early or partial results. Treat as a signal, not a conclusion.

AI Summary

FedLBW is a proposed aggregation algorithm for federated learning that weights client updates by inverse validation loss rather than dataset size. In simulation experiments on image classification benchmarks, it reports higher accuracy and faster convergence than six baseline methods, with claimed improvements up to 7.6% on CIFAR-10 under extreme non-IID conditions and resilience to client dropout. The evidence is computational and exploratory, suitable for algorithm development but insufficient to support deployment decisions without validation on real networks and heterogeneous data.

What was studied

Algorithm development with simulation-based experimental validation. Simulated federated learning environments with non-IID distributed data and varying client dropout rates; no real-world network deployment or human subjects.. Intervention: FedLBW: loss-based weighting aggregation method using inverse validation loss on server proxy dataset. Compared with: FedAvg, FedAvgM, FedProx, FedNova, FedLAW, FedDkw.

Key findings

FedLBW achieves up to 7.6% higher accuracy on CIFAR-10 compared to baselines in extreme non-IID cases Algorithm demonstrates faster convergence and improved accuracy across FashionMNIST (CNN), CIFAR-10 (ResNet-18), and CIFAR-100 (ResNet-34) FedLBW shows exceptional resilience to increasing dropout probabilities, maintaining significantly higher accuracy in challenging conditions

Safety

Safety was not reported in the material analysed. Check the source before drawing any conclusion about harm.

Who this may help

The source did not state who this applies to in practice.

Plain-language summary

Simulation-based algorithmic study demonstrating improved federated learning convergence on benchmark datasets; lacks clinical validation, real-world deployment evidence, or comparison to state-of-the-art methods in applied settings.

Study design

As stated by the source record.

MethodsExtensive experiments conducted across multiple image classification datasets (FashionMNIST, CIFAR-10, CIFAR-100) using different model architectures (CNN, ResNet-18, ResNet-34). The proposed FedLBW algorithm assigns weights to client updates proportional to inverse validation loss computed on a server-side proxy dataset, contrasting with size-based weighting in baseline methods. Performance compared against six baseline algorithms (FedAvg, FedAvgM, FedProx, FedNova, FedLAW, FedDkw) under varying non-IID data distributions and client dropout conditions. No specification of hyperparameter sele…

DesignAlgorithm development with simulation-based experimental validation

PopulationSimulated federated learning environments with non-IID distributed data and varying client dropout rates; no real-world network deployment or human subjects.

InterventionFedLBW: loss-based weighting aggregation method using inverse validation loss on server proxy dataset

ComparatorFedAvg, FedAvgM, FedProx, FedNova, FedLAW, FedDkw

Reported figures

Quoted from the source exactly as published.

Accuracy improvement on CIFAR-10 (extreme non-IID)7.6%

Caveats

Source does not report absolute accuracy values, confidence intervals, or statistical significance testing for any result
Number of clients, dropout rates tested, and per-client dataset sizes not specified
Comparison limited to federated learning baselines; no comparison to centralized learning or recent distributed optimization methods
Applicability to real wireless networks with actual communication constraints, bandwidth limits, and latency not evaluated

Evidence assessment

Graded across the dimensions that decide whether you should act, each from what the source actually supports. There is no single score, and where a dimension was not assessed it says so.

Study design

moderate

Controlled computational experiments across multiple datasets; not randomized or blinded.

Directness

limited

Benchmark datasets (FashionMNIST, CIFAR-10, CIFAR-100) used; applicability to real wireless networks unclear.

Consistency

moderate

Improvements demonstrated across three datasets; comparison to six baseline algorithms.

Risk of bias

insufficient

No blinding, randomization, or preregistration reported; hyperparameter tuning approach not disclosed.

Sample adequacy

insufficient

Client count, dropout rates, and dataset sizes per client not specified in source text.

Source abstract

Federated Learning (FL) enables collaborative machine learning (ML) across distributed clients while preserving privacy. However, efficient model convergence in FL remains challenging, especially in wireless networks where non-independent and identically distributed (non-IID) data and frequent client dropouts are common. Traditional FL algorithms, such as FedAvg, rely solely on dataset size to weight client updates. This introduces biases towards clients with larger datasets and makes the process sensitive to non-IID data, outliers, and client dropouts. To address these challenges, we propose Federated Learning with Loss-Based Weighting (FedLBW), a novel aggregation method that assigns each client's update a weight proportional to the inverse of its validation loss, computed using a small proxy dataset on the server, rather than its dataset size. This ensures that lower-loss models exert greater influence during aggregation, prioritizing the most reliable updates and boosting overall performance. Through extensive experiments across multiple datasets, including FashionMNIST (CNN), CIFAR-10 (ResNet-18), and CIFAR-100 (ResNet-34), we demonstrate that FedLBW achieves higher accuracy and faster convergence compared to baseline algorithms such as FedAvg, FedAvgM, FedProx, FedNova, FedLAW and FedDkw, with notable improvements of up to 7.6 % higher accuracy on CIFAR-10 in extreme non-IID cases. Moreover, FedLBW showcases exceptional resilience to increasing dropout probabilities, consistently maintaining significantly higher accuracy even in challenging conditions. These results establish FedLBW as an effective and resilient solution for FL in wireless network environments, offering marked improvements in model accuracy, convergence speed, and robustness to non-IID data and client dropouts.

Follow from this

Taken from the source record, never inferred. Follow any of these and new work involving them reaches your briefing.

Majid KundrooTinku SinghTaehong Kim