arxiv: 2605.08121 · v1 · submitted 2026-04-28 · 💻 cs.DC · cs.LG

Recognition: no theorem link

Performance and Energy Trade-Off Analysis of Hierarchical Federated Learning for Plant Disease Classification

Athanasios Papanikolaou , Athanasios Tziouvaras , Pavlos Stoikos , Apostolos Xenakis , Shameem A Puthiya Parambath , George Floros , Enrica Zereik , Ivan Petrovic

show 1 more author

Fabio Bonsignorio

Authors on Pith no claims yet

Pith reviewed 2026-05-12 01:07 UTC · model grok-4.3

classification 💻 cs.DC cs.LG

keywords hierarchical federated learningplant disease classificationenergy efficiencyperformance trade-offsconvolutional neural networksfederated aggregation strategiesIoT environmentsprecision agriculture

0 comments

The pith

Hierarchical federated learning for plant disease classification reveals distinct performance-energy trade-offs across model and aggregator choices.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines hierarchical federated learning architectures for classifying plant diseases in distributed IoT settings. It emphasizes the balance between model accuracy and energy consumption by exploring various convolutional neural networks and aggregation strategies. A dedicated optimization framework is presented to assess these choices under different constraints. Results highlight that select combinations deliver strong diagnostic performance while lowering overall system resource use.

Core claim

Organizing clients via intermediate layers in a hierarchical federated setup cuts communication and computation overhead. Testing EfficientNet-B0, ResNet-50, and MobileNetV3-Large with FedAvg, FedProx, and FedAvgM shows unique accuracy versus energy profiles for each pairing, where some options keep diagnostic accuracy competitive and cut resource demands substantially.

What carries the argument

Hierarchical federated architecture with intermediate aggregation layers that organizes distributed clients to reduce overhead, paired with the power- and energy-aware optimization framework for configuration selection.

If this is right

Configurations can be selected to match specific deployment energy or performance needs.
System efficiency improves for large-scale IoT-based agricultural monitoring.
The evaluation method supports adapting to varying constraints in distributed learning.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Scalability of AI-driven precision agriculture increases as energy barriers decrease.
Similar hierarchical designs could apply to other energy-sensitive edge AI tasks beyond farming.

Load-bearing premise

The hierarchical architecture reduces communication and computational overhead without materially harming model convergence or final accuracy under the tested deployment constraints.

What would settle it

Measurement in a real-world IoT deployment showing that a high-performing configuration either consumes more energy than expected or achieves lower diagnostic accuracy than non-hierarchical baselines.

Figures

Figures reproduced from arXiv: 2605.08121 by Apostolos Xenakis, Athanasios Papanikolaou, Athanasios Tziouvaras, Enrica Zereik, Fabio Bonsignorio, George Floros, Ivan Petrovic, Pavlos Stoikos, Shameem A Puthiya Parambath.

**Figure 1.** Figure 1: The hierarchical FL concept, where a problem is broken down into concrete subtasks, each one solvable by a DNN model. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Figure 2: EfficientNet-B0/ResNet-50/MobileNetV3-Large loss over epochs, using fedavg, fedprox and fedavgm. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

read the original abstract

Early detection of plant diseases is critical for improving crop productivity, while it also facilitates the foundations of precision agriculture. Recent advances in distributed deep learning have enabled plant disease classification models to be trained across geographically distributed agricultural sensing infrastructures. However, deploying such systems in large-scale Internet of Things (IoT) environments, introduces significant challenges related to computational cost, energy consumption, and system efficiency. In this paper, we present a design-space exploration of hierarchical federated learning architectures for plant disease classification, with a particular focus on the trade-offs between predictive performance and energy efficiency. We further introduce a power- and energy-aware optimization framework that enables the systematic evaluation and selection of model-aggregator configurations under varying deployment constraints. The hierarchical federated architecture organizes distributed clients through intermediate aggregation layers, reducing communication and computational overhead. We evaluate multiple convolutional neural network architectures, including EfficientNet-B0, ResNet-50, and MobileNetV3-Large, in combination with different federated aggregation strategies such as FedAvg, FedProx, and FedAvgM. Experimental results demonstrate that different model-aggregator combinations exhibit distinct performance-energy trade-offs. Consequently, we highlight configurations that achieve competitive diagnostic accuracy and significantly reduce system resource requirements.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This applies standard hierarchical federated learning to plant disease images and maps energy-accuracy trade-offs across a few CNNs and aggregators, but the hierarchy benefit is not isolated from other factors.

read the letter

The paper runs a design-space exploration of hierarchical federated learning for plant disease classification. It organizes clients through intermediate layers, tests EfficientNet-B0, ResNet-50, and MobileNetV3-Large with FedAvg, FedProx, and FedAvgM, and tracks both diagnostic accuracy and energy use under IoT-style constraints. They also describe a power-aware framework for picking configurations based on deployment limits. That systematic comparison of known pieces in a new domain is the main contribution and gives practitioners some concrete options for balancing performance against resource cost in agricultural sensing setups. The experiments appear to identify combinations that hold accuracy while cutting energy draw, which matches the practical goal stated in the abstract. The central soft spot is the absence of a flat federated learning control. The abstract credits the intermediate aggregation layers for lower overhead, yet the reported runs vary only models and aggregators inside the hierarchical structure. Without matched flat runs on the same data partitions, round counts, and measurement protocol, it is hard to tell how much of the reported savings comes from the hierarchy itself versus lighter models or different training dynamics. The abstract also omits any numbers, dataset sizes, error bars, or statistical tests, so the full paper must supply those to make the trade-off claims verifiable. This work is aimed at engineers and applied researchers who need guidance on energy-constrained federated learning in farming IoT rather than theorists looking for new algorithms. It is honest empirical work that engages the right constraints, so it deserves a serious referee even though the hierarchy attribution needs tightening.

Referee Report

1 major / 1 minor

Summary. The paper presents a design-space exploration of hierarchical federated learning architectures for plant disease classification in IoT environments. It evaluates combinations of CNN models (EfficientNet-B0, ResNet-50, MobileNetV3-Large) and aggregation strategies (FedAvg, FedProx, FedAvgM) within a hierarchical client-intermediate-server structure, introduces a power- and energy-aware optimization framework for selecting configurations under deployment constraints, and reports experimental results showing distinct performance-energy trade-offs, with the hierarchical organization claimed to reduce communication and computational overhead while preserving competitive diagnostic accuracy.

Significance. If the results hold and the hierarchical structure can be shown to be the primary driver of resource savings, the work could inform practical deployment of distributed ML systems in resource-constrained agricultural sensing applications. The emphasis on joint performance-energy evaluation and the optimization framework represent a useful contribution to the intersection of FL and edge/IoT systems for domain-specific tasks like plant disease detection.

major comments (1)

Abstract: The central claim credits the hierarchical architecture with reducing communication and computational overhead ('The hierarchical federated architecture organizes distributed clients through intermediate aggregation layers, reducing communication and computational overhead'). However, the described experiments vary only CNN architectures and aggregators inside the hierarchical structure and provide no matched non-hierarchical (flat) FL baseline with identical data partitions, round counts, and energy measurement protocols. This prevents isolation of the hierarchy's contribution from model size or convergence effects, directly undermining the attribution of overhead reduction to the hierarchical organization and the paper's weakest assumption.

minor comments (1)

Abstract: The abstract asserts that experiments demonstrate trade-offs and that certain configurations achieve competitive accuracy with reduced resources, yet reports no quantitative metrics, error bars, dataset sizes, or statistical tests. Adding key numerical results would make the summary of findings more informative and verifiable.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their insightful comments on our paper. We address the major concern point by point below.

read point-by-point responses

Referee: Abstract: The central claim credits the hierarchical architecture with reducing communication and computational overhead ('The hierarchical federated architecture organizes distributed clients through intermediate aggregation layers, reducing communication and computational overhead'). However, the described experiments vary only CNN architectures and aggregators inside the hierarchical structure and provide no matched non-hierarchical (flat) FL baseline with identical data partitions, round counts, and energy measurement protocols. This prevents isolation of the hierarchy's contribution from model size or convergence effects, directly undermining the attribution of overhead reduction to the hierarchical organization and the paper's weakest assumption.

Authors: We acknowledge that the absence of a direct comparison to a non-hierarchical (flat) FL baseline limits our ability to quantitatively isolate the contribution of the hierarchical structure to the observed reductions in communication and computational overhead. Our study primarily explores the design space of CNN models and aggregation strategies within a hierarchical FL framework for plant disease classification, emphasizing performance-energy trade-offs. The statement in the abstract reflects the general architectural advantage of hierarchical FL, where intermediate aggregation layers reduce the volume of data transmitted to the central server, as established in the FL literature. However, to strengthen the manuscript, we will revise the abstract to more precisely attribute the overhead reduction to the hierarchical design principles rather than claiming it as a direct result of our experiments. We will also add a paragraph in the discussion section acknowledging this as a limitation and outlining how future work could include flat baselines for comparison. This constitutes a partial revision since adding full experimental baselines would require substantial additional resources and time. revision: partial

Circularity Check

0 steps flagged

No significant circularity: experimental results are direct measurements, not derived quantities

full rationale

The paper conducts a design-space exploration of hierarchical federated learning for plant disease classification by evaluating specific CNN architectures (EfficientNet-B0, ResNet-50, MobileNetV3-Large) paired with aggregators (FedAvg, FedProx, FedAvgM) under a fixed hierarchical client-intermediate-server structure. All reported outcomes are presented as empirical measurements of accuracy, communication overhead, and energy consumption from these runs. No equations, fitted parameters, or predictions are defined in terms of earlier results; the hierarchical organization is the experimental condition itself rather than a quantity derived from prior fitted values or self-citations. The central claims rest on direct experimental comparison within the tested configurations, with no load-bearing derivation chain that reduces to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical derivations, free parameters, or invented entities appear in the provided abstract; the work is framed as an empirical design-space exploration.

pith-pipeline@v0.9.0 · 5563 in / 987 out tokens · 27149 ms · 2026-05-12T01:07:46.680356+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages

[1]

Big data in smart farming – a review,

S. Wolfert, L. Ge, C. Verdouw, and M.-J. Bogaardt, “Big data in smart farming – a review,”Agricultural Systems, vol. 153, pp. 69–80, 2017

work page 2017
[2]

Advances in plant disease detection and monitoring: From traditional assays to in-field diagnostics,

I. Buja, E. Sabella, A. G. Monteduro, M. S. Chiriac `o, L. De Bellis, A. Luvisi, and G. Maruccio, “Advances in plant disease detection and monitoring: From traditional assays to in-field diagnostics,”Sensors, vol. 21, no. 6, p. 2129, 2021

work page 2021
[3]

Internet of things (IoT) and machine learning model of plant disease prediction–blister blight for tea plant,

Z. Liu, R. N. Bashir, S. Iqbal, M. M. A. Shahid, M. Tausif, and Q. Umer, “Internet of things (IoT) and machine learning model of plant disease prediction–blister blight for tea plant,”IEEE Access, vol. 10, pp. 44 934–44 944, 2022

work page 2022
[4]

En- abling all in-edge deep learning: A literature review,

P. Joshi, M. Hasanuzzaman, C. Thapa, H. Afli, and T. Scully, “En- abling all in-edge deep learning: A literature review,”IEEE Access, vol. 11, pp. 3431–3460, 2023

work page 2023
[5]

Internet of things and wireless sensor networks for smart agriculture applications: A survey,

M. N. Mowla, N. Mowla, A. S. Shah, K. M. Rabie, and T. Shongwe, “Internet of things and wireless sensor networks for smart agriculture applications: A survey,”IEEE Access, vol. 11, 2023

work page 2023
[6]

Edge-enabled smart agriculture framework: Integrating IoT, lightweight deep learning, and agentic ai for context-aware farming,

M. U. Tariq, S. M. Saqib, T. Mazhar, M. A. Khan, T. Shahzad, and H. Hamam, “Edge-enabled smart agriculture framework: Integrating IoT, lightweight deep learning, and agentic ai for context-aware farming,”Results in Engineering, vol. 28, p. 107342, 2025

work page 2025
[7]

Deploying edge computing nodes for large-scale iot: A diversity aware approach,

Z. Zhao, G. Min, W. Gao, Y . Wu, H. Duan, and Q. Ni, “Deploying edge computing nodes for large-scale iot: A diversity aware approach,” IEEE Internet of Things Journal, vol. 5, no. 5, pp. 3606–3614, 2018

work page 2018
[8]

From federated learning to federated neural architecture search: a survey,

H. Zhu, H. Zhang, and Y . Jin, “From federated learning to federated neural architecture search: a survey,”Complex & Intelligent Systems, vol. 7, no. 2, pp. 639–657, 2021

work page 2021
[9]

Distributed deep learning in iot sensor network for the diagnosis of plant diseases,

A. Papanikolaou, A. Tziouvaras, G. Floros, A. Xenakis, and F. Bon- signorio, “Distributed deep learning in iot sensor network for the diagnosis of plant diseases,”Sensors, vol. 25, no. 24, 2025

work page 2025
[10]

Communication-Efficient Learning of Deep Networks from Decen- tralized Data,

B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y. Arcas, “Communication-Efficient Learning of Deep Networks from Decen- tralized Data,” inProceedings of the 20th International Conference on Artificial Intelligence and Statistics, ser. Proceedings of Machine Learning Research, A. Singh and J. Zhu, Eds., vol. 54. PMLR, 20–22 Apr 2017, pp. 1273–1282

work page 2017
[11]

Federated optimization in heterogeneous networks,

T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V . Smith, “Federated optimization in heterogeneous networks,”Proceedings of Machine learning and systems, vol. 2, pp. 429–450, 2020

work page 2020
[12]

Adaptive federated optimization,

S. Reddi, Z. Charles, M. Zaheer, Z. Garrett, K. Rush, J. Kone ˇcn`y, S. Kumar, and H. B. McMahan, “Adaptive federated optimization,” arXiv preprint arXiv:2003.00295, 2020

work page arXiv 2003
[13]

EfficientNet: Rethinking model scaling for convolutional neural networks,

M. Tan and Q. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” inProceedings of the 36th Interna- tional Conference on Machine Learning, ser. Proceedings of Machine Learning Research, K. Chaudhuri and R. Salakhutdinov, Eds., vol. 97. PMLR, 09–15 Jun 2019, pp. 6105–6114

work page 2019
[14]

Deep residual learning for image recognition,

K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778

work page 2016
[15]

Search- ing for mobilenetv3,

A. Howard, M. Sandler, B. Chen, W. Wang, L.-C. Chen, M. Tan, G. Chu, V . Vasudevan, Y . Zhu, R. Pang, H. Adam, and Q. Le, “Search- ing for mobilenetv3,” in2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 1314–1324

work page 2019
[16]

Plantdoc: A dataset for visual plant disease detection,

D. Singh, N. Jain, P. Jain, P. Kayal, S. Kumawat, and N. Batra, “Plantdoc: A dataset for visual plant disease detection,” inProceedings of the 7th ACM IKDD CoDS and 25th COMAD, ser. CoDS COMAD 2020, 2020, p. 249–253

work page 2020