Asynchronous Probability Ensembling for Federated Disaster Detection

Emanuel Teixeira Martins; Rodrigo Moreira; Larissa Ferreira Rodrigues Moreira; Rodolfo S. Villaça; Augusto Neto; Flávio de Oliveira Silva

cs.LGApr 15, 2026

Asynchronous Probability Ensembling for Federated Disaster Detection

Authors: Emanuel Teixeira Martins, Rodrigo Moreira, Larissa Ferreira Rodrigues Moreira, Rodolfo S. Villaça, Augusto Neto, Flávio de Oliveira Silva

Organizations: Federal University of Espirito Santo (UFES), Vitória-ES, Brazil · Federal University of Rio Grande do Norte (UFRN), Natal-RN, Brazil · University of Minho (UMinho), Braga, Portugal

Abstract

Quick and accurate emergency handling in Disaster Decision Support Systems (DDSS) is often hampered by network latency and suboptimal application accuracy. While Federated Learning (FL) addresses some of these issues, it is constrained by high communication costs and rigid synchronization requirements across heterogeneous convolutional neural network (CNN) architectures. To overcome these challenges, this paper proposes a decentralized ensembling framework based on asynchronous probability aggregation and feedback distillation. By shifting the exchange unit from model weights to class-probability vectors, our method maintains data privacy, reduces communication requirements by orders of magnitude, and improves overall accuracy. This approach enables diverse CNN designs to collaborate asynchronously, enhancing disaster image identification performance even in resource-constrained settings. Experimental tests demonstrate that the proposed method outperforms traditional individual backbones and standard federated approaches, establishing a scalable and resource-aware solution for real-time disaster response.

Figures & tables

Work	Field	FEL (prob.)	Client Selection	Async. Training	Heterogeneous Models	AI Models
[ 7 ]	Healthcare					Multiple CNNs
[ 9 ]	Healthcare					ResNet-20
[ 10 ]	Adversarial FL					Multiple CNNs
[ 11 ]	Smart Grids					LSTM, CNN, Transformer
[ 12 ]	Infrastructure Monitoring					EfficientNet
[ 13 ]	Aviation					LSTM, CNN

TABLE I: Comparison with related work. The symbols denote full support ( ), partial or adaptive support ( ), and no support ( ).

P^{(i)}(x)=f_{i}(x),\hskip 8.19447pt\forall x\in X_{\mathrm{ref}}

Algorithm 1 Knowledge Distillation with a Shared Reference Set.

Parameter	Value
Population size	40
Generations	100
Selection strategy	Elitism (top 5 preserved)
Crossover	One-point
Mutation rate ( $p_{m}$ )	0.3 (Gaussian noise)
Diversity injection	Every 10 generations

TABLE II: Genetic Algorithm Parameters.

Parameter	Value
Swarm size	20
Iterations	100
Inertia coefficient ( $\omega$ )	0.7
Cognitive factor ( $c_{1}$ )	1.5
Social factor ( $c_{2}$ )	1.5
Fitness function	Validation accuracy

TABLE III: Particle Swarm Optimization Parameters.

Hyperparameter	Distribution	Search Space
LR ( LR )	Log-Uniform	$1\times 10^{-5}$ , $1\times 10^{-3}$
Batch Size	Categorical	$\{16,32,64\}$
Optimizer	Categorical	$\{\text{Adam, SGD}\}$

TABLE IV: Hyperparameter Search Space.

Model	Accuracy		Macro-F1
Model	Centralized	FL	Centralized	FL
ResNet	0.9798	0.9813	0.9668	0.9681
EfficientNet	0.9783	0.9813	0.9633	0.9633
MobileNetV2	0.9659	0.9720	0.9431	0.9533
MobileNetV3	0.9666	0.9767	0.9407	0.9584
SqueezeNet	0.9465	0.9557	0.9087	0.9270

TABLE V: Centralized vs FL on AIDER : Accuracy and Macro-F1.

Fig. 2: Distribution of accuracy gains achieved by the ensemble.

Fig. 3: Cumulative distribution of ensemble gains over the best individual model.

Combination	Accuracy	Macro F1	Weighted F1
EfficientNet + MobileNetV2 + ResNet	0.98216	0.96990	0.98204
MobileNetV2 + MobileNetV3 + SqueezeNet	0.97285	0.95306	0.97248

TABLE VI: Comparison of ensemble combinations.

Model	Ensemble	Federated
SqueezeNet	149,603	8,855,868
ResNet	149,231	255,651,996
MobileNetV3	149,231	18,421,932
MobileNetV2	151,184	27,173,916
EfficientNet	148,649	48,672,804

TABLE VII: Network consumption (bytes transmitted) for Ensemble vs. FL

Federated Learning

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

See all 943 papers

Reviving Stale Updates: Data-Free Knowledge Distillation for Asynchronous Federated Learning
Nov 1, 2025Baris Askin, Holger R. Roth, Zhenyu Sun +3Asynchronous Federated Learning Federated Learning Aggregation
Carnegie Mellon University · NVIDIA · Northwestern University
Optimized Federated Knowledge Distillation with Distributed Neural Architecture Search
May 20, 2026Chaimaa Medjadji, Sylvain Kubler, Yves Le Traon +3Communication-Efficient Distributed Training Federated Knowledge Distillation
Interdisciplinary Centre for Security, Reliability and Trust (SnT), University of Luxembourg, Luxembourg · Blekinge Institute of Technology, Sweden · ADSLabs, Umea University, Sweden
On the Push-Based Asynchronous Federated Learning: A Bias-Correction Aggregation Approach
May 24, 2026Jiahui Bai, Hai Dong, A. K. QinMulti-Agent Coordination Federated Learning Aggregation
School of Computer Technologies, RMIT University Melbourne, VIC, Australia · School of Science, Computing and Engineering Technologies, Swinburne University of Technology Hawthorn, VIC, Australia

Nov 1, 2025J/K move · Enter open · S save

Reviving Stale Updates: Data-Free Knowledge Distillation for Asynchronous Federated Learning

Baris Askin, Holger R. Roth, Zhenyu Sun, Carlee Joe-Wong+2

Carnegie Mellon University · NVIDIA · Northwestern University

Asynchronous Federated Learning Federated Learning Aggregation

Open paper AskPDF arXiv

Federated learning (FL) enables collaborative model training across distributed clients without sharing raw data, yet its scalability is limited by synchronization overhead. Asynchronous federated learning (AFL) alleviates this issue by allowing clients to communicate independently, thereby improving wall-clock efficiency in large-scale, hardware-heterogeneous environments. However, asynchrony introduces updates computed on outdated global models (staleness) that can destabilize optimization and hinder convergence. We propose FedRevive, an AFL framework that revives stale updates through data-free knowledge distillation (DFKD). FedRevive integrates parameter-space aggregation with a server-side DFKD process that transfers knowledge from stale client updates to the current global model without access to data. A meta-learned generator synthesizes pseudo-samples for multi-teacher distillation. A hybrid aggregation scheme combining raw with DFKD updates effectively mitigates staleness while retaining AFL scalability. Experiments on vision and text benchmarks show that FedRevive achieves faster training by up to 29.7% and higher final accuracy by up to 11.4 percentage points than baselines.

Abstract

Federated Learning

Explore similar work

Reviving Stale Updates: Data-Free Knowledge Distillation for Asynchronous Federated Learning

Optimized Federated Knowledge Distillation with Distributed Neural Architecture Search

On the Push-Based Asynchronous Federated Learning: A Bias-Correction Aggregation Approach