Pith. sign in

REVIEW 3 major objections 7 minor 61 references

Loss-Guided Model Sharing and Local Learning Correction in Decentralized Federated Learning for Crop Disease Classification

T0 review · 3 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read A server-free federated network can be steered by validation loss alone: only low-loss models circulate, and their validation loss, weighted by λ, is folded into each client's local loss.

desk verdict The paper's central 'local learning correction' adds a constant to the loss and is never optimized, so λ cannot work; the reported gains must come from an unstated mechanism or error. read the letter →

arxiv 2505.23063 v1 pith:K6PAAW4U submitted 2025-05-29 cs.LG

classification cs.LG
keywords decentralizedfederatedlearningvalidationlossmodelsharinglocalcorrectioncropdiseaseclassificationpeer-to-peeragriculturalimagedataaveraging
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Federated learning normally leans on a central server to aggregate models, which creates a single point of failure. This paper proposes a decentralized, peer-to-peer alternative for crop-disease classification in which validation loss does double duty: each client shares its model only with neighbors that have higher validation loss, and the validation loss of received models, weighted by a hyperparameter λ, is added to the client's local loss to correct its learning. On grape, apple, and corn leaf datasets, the authors report that positive λ improves accuracy, convergence speed, and stability across ResNet50, VGG16, and ViT_B16, with best F1-scores reaching 99%, and that validation loss works better than training loss as the correction signal. If true, this would give privacy-sensitive agricultural users a server-free way to train better classifiers without sharing raw images.

What carries the argument

The two load-bearing mechanisms are the sharing criterion—neighbor $j$ receives client $k$'s model only if $\mathrm{Loss}_{val,j}^t > \mathrm{Loss}_{val,k}^t$—and the adjusted local loss $\mathrm{Loss}_k^t = \mathrm{Local\_loss}_k^t + \lambda \cdot \mathrm{Loss}_{received}^t$, where $\mathrm{Loss}_{received}^t$ is the average validation loss of the best received models. The sharing rule determines which models circulate (only the strongest), while the adjusted loss is intended to correct local training by injecting peer performance into the next round, with λ controlling how much external influence each client accepts. Aggregation uses FedAvg, defined as plain averaging of model weights, to combine the local model with received best models.

What would settle it

Run the protocol twice from the same random seed with λ=0 and λ>0, logging gradients and aggregations; if the adjusted loss never enters the optimizer, the two runs must be identical, which would show the reported λ effects come from outside the described algorithm.

Watch

Extended reading notes

Core claim

The central claim is that a decentralized federation can be steered entirely by validation loss, with no central server and no raw data exchange. In the proposed protocol, at each round every client computes its own validation loss; a client sends its model only to neighbors with strictly higher validation loss, so only low-loss models travel. Each client aggregates its local model with the best received models using FedAvg, then adjusts its local loss by adding the averaged validation loss of the received models, weighted by a hyperparameter λ in {0, 0.25, 0.5, 0.75}. The authors claim this combination improves accuracy and convergence, reduces communication cost through selective sharing, and improves robustness across heterogeneous data. Their experiments on three leaf-disease datasets with ResNet50, VGG16, and ViT_B16 show positive λ generally raising F1-score and lowering variance, with ResNet50 reaching 99% F1-score; they also find validation loss is a better correction signal than training loss, especially for ResNet50.

Load-bearing premise

The results require that the λ-weighted validation loss of the received model actually change the local parameter updates, but in the algorithm and loss as written it is a constant with respect to the local model's parameters and therefore has zero gradient.

Editorial extensions

If this is right

  • Farmers or cooperatives could train classifiers across farms without a central server, exchanging only model weights and scalar losses rather than leaf images.
  • Selective sharing by validation loss lowers communication: with VGG16, sharing one best model gives nearly the same F1-score as sharing five, so less traffic can be enough.
  • The λ parameter gives a per-architecture dial between local-data fidelity and global generalization, with VGG16 and ViT_B16 preferring small λ and ResNet50 preferring large λ.
  • Using validation loss rather than training loss as the correction signal yields more stable convergence and better generalization, which should help when clients' local data are not balanced.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A control experiment that replaces the received validation loss with a random scalar would isolate whether the λ-dependent gains come from the written loss correction or from the model-sharing and aggregation strategy alone.
  • Making the correction differentiable—for example, penalizing divergence from the received model's predictions—would give the proposed loss a real gradient and allow it to steer local training as intended.
  • The heterogeneity claim could be tested more stringently by assigning each client a different crop subset rather than duplicated splits, a non-IID setting the authors flag as future work.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. The paper proposes a decentralized federated learning (DFL) framework for crop disease classification in which clients share models with peers whose validation loss is higher, aggregate received models via FedAvg, and attempt to "correct" local training by adding a validation-loss term from the received model to the local loss, weighted by a hyperparameter λ (Eq. 5). The authors report experiments on PlantVillage subsets using ResNet50, VGG16, and ViT_B16 with 6 or 18 clients, one or five shared models, and λ values from 0 to 0.75. They claim improved accuracy, faster convergence, and better generalization and robustness compared to centralized FL and to λ = 0.

Significance. If the proposed mechanism worked, the framework would offer a simple, privacy-preserving way for distributed agricultural clients to improve classifiers without a central server. The paper deserves credit for testing multiple architectures, client counts, and sharing configurations, and for comparing against prior work. However, the central contribution is undermined by an internal inconsistency: the "local learning correction" term in Eq. (5) is a constant with respect to the client's current parameters and is never used in the optimizer updates shown in Algorithm 1. Consequently, λ cannot influence training as described, and the λ-dependent performance differences in Tables III-VI are unexplained. The experimental protocol also selects λ and the number of shared models using final test performance on the same partitions, compromising generalization claims. These issues are load-bearing, so the central claims are not supported.

major comments (3)
  1. [Section III-A5 and Algorithm 1] Equation (5) defines Loss^k = Local_loss^k + λ * Loss_received_model, but Loss_received_model is computed from the received (aggregated) model's validation data, whose parameters are fixed at the start of the next local-training phase. As a function of the current client's local parameters, its gradient is zero, so adding it to the local loss cannot alter any gradient-based parameter update. Algorithm 1 (lines 10-12) confirms that the local update uses only the ordinary batch loss L(A_{t-1}^k; b), not the adjusted loss from Eq. (5). The abstract's claimed accuracy and convergence gains are explicitly attributed to this λ-weighted correction, so the central mechanism is unsupported as written. The λ-dependent results in Tables III-VI must therefore come from an unstated mechanism, randomness, or evaluation artifact, not from the described method.
  2. [Section IV-A.3 and Section IV-B (Tables III-VI)] The same 20% held-out partition is used both as the validation set for selecting shared models (via Loss_val in Eq. 3) and as the test set for reporting final accuracy and F1-score. In addition, λ and the number of shared models are chosen after comparing final performance on these same tables, so the reported "optimal" configurations are fitted to the test data. This circularity means the reported gains over λ = 0 and the cross-architecture conclusions do not establish generalization to unseen data.
  3. [Section IV-B, Table VII] The comparison with prior work is not controlled. The authors compare their results with [28], [56], and [57] using different numbers of clients, different data partitions, and different training protocols, but provide no details on how the literature results were obtained under the same conditions. The claim that the proposed framework "outperforms" these methods is therefore not supported by the evidence presented.
minor comments (7)
  1. [Section III-B] "tree pre-trained models" should read "three pre-trained models."
  2. [Conclusion] "non-IDI" should be "non-IID."
  3. [Throughout] Algorithm 1 and Section III-A use "test" and "validation" interchangeably for the same 20% partition; the terminology should be made consistent.
  4. [Table V] The rows for VGG16 and ViT_B16 are identical for all λ values, which appears to be a transcription error and should be corrected.
  5. [Table IV] The column header for VGG16 repeats "F1-Score Accuracy F1-Score," making the column assignment ambiguous.
  6. [Section IV-A] The paper does not state how the reported standard deviations are computed (across clients, rounds, or runs) or how many independent runs each configuration used.
  7. [Table II] The table lists "Epoch/Client" as 6 and 18, but the text and Algorithm 1 state one local epoch per client; this should be clarified.

Circularity Check

2 steps flagged · score 6.0 of 10

The proposed local learning correction is a constant offset: the λ-weighted received-model loss in Eq. (5) has zero gradient w.r.t. the client's own parameters, and Algorithm 1 never optimizes Eq. (5), so the reported λ-dependent gains are not derived from the stated mechanism; λ is also selected on the evaluation set.

  1. other [Section III-A-5 'Update local loss function', Eqs. (5)-(6); Algorithm 1 lines 12 and 25-27]
    "Loss^t_k = Local loss^t_k + λ · Loss^t_received model ... The new value of the loss function obtained will be used in the next cycle in the local learning of the model to correct or readjust the learning ... Algorithm 1 line 12: L^t_k ← L^t_k − η∇L(A^{t−1}_k ; b)."

    By Eq. (6), Loss^t_received_model is the validation loss of the best models received from neighbors. Those models' parameters are fixed when client k begins its next local-training loop, so ∂/∂θ_k(λ·Loss^t_received_model) = 0. Adding this constant to Local loss^t_k cannot change any gradient-based update. Algorithm 1's only local update is line 12, which uses the ordinary batch loss L(A^{t−1}_k; b), never the adjusted loss from Eq. (5). The proposed 'local learning correction' is therefore identical, by construction, to the λ = 0 optimization problem.

  2. fitted input called prediction [Section III-A-5 (λ selection) and Section IV-B-1 'Impact of the λ parameter']
    "we empirically chose and tested four specific values of λ ∈ [0, 1], namely : 0, 0.25, 0.50 and 0.75, in order to assess their impact on model performance and to determine, after an analysis of the results obtained for each configuration, which of the λ values is optimal in our case."

    The parameter λ is selected by analyzing final performance on the same 20% partition that Eq. (3) calls D^test_k and that Table III uses to report accuracy and F1-score. The paper then presents the best λ for each architecture as evidence that the validation-loss correction improves generalization. Because the evaluation metric is used as the selection criterion, the reported 'optimal' λ and its accuracy are fitted to the test labels rather than independently predicted; the test set does double duty as both validation and final evaluation set.

full rationale

The central contribution of the paper is the 'local learning correction' in which the validation loss of received models is added to the local loss with a λ weight. This mechanism is inert as written: Eq. (5) adds a constant with respect to the client's own parameters, and Algorithm 1 never optimizes Eq. (5). Therefore the proposed correction reduces by construction to the ordinary local loss, and the λ-dependent improvements in Tables III-VI are not explained by the paper's stated method. This is a specific algebraic reduction, not a vague concern about reproducibility. The additional selection of λ on the same test partition used for final reporting compounds the issue, making the reported optimal configurations fitted rather than predicted. The paper does not rely on load-bearing self-citation chains or imported uniqueness theorems, and its external comparisons are against PlantVillage benchmarks, so the score is not 8-10; however, because the central claim is unsupported by the paper's own equations, a score of 6 is appropriate.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new particles, forces, or entities. The free parameters are lambda and the number of shared models, both selected via test performance. The most consequential axiom is the unstated assumption that the adjusted loss is used in training, which is contradicted by the provided algorithm.

free parameters (2)
  • lambda (weighting parameter) = 0.25, 0.50, 0.75 (optimal chosen per model)
    The paper tests four lambda values and selects the best based on test performance on PlantVillage. This is a free parameter fitted to the evaluation data.
  • Number of shared models (n_best) = 1 or 5
    The paper experiments with sharing 1 or 5 models and interprets the impact. The choice of 5 is evaluated on the same test data; no separate validation set is used for selecting this hyperparameter.
assumptions (3)
  • domain assumption The validation loss of a model computed on the sender's test set is a meaningful indicator of model quality for other clients.
    This underlies the sharing criterion in Phase 3 (if Loss_val,j > Loss_val,k then k shares with j). In heterogeneous data settings, a low loss on one client's test set does not guarantee good performance on another client's distribution.
  • ad hoc to paper The adjusted loss in Equation (5) is the objective minimized during local training.
    Algorithm 1 line 12 updates the model using the batch loss L(A^{t-1}_k; b), not the adjusted loss from Equation (5). The paper does not show how the adjusted loss enters the optimizer. This axiom is necessary for the claimed correction mechanism to have any effect, but it does not follow from the algorithm.
  • domain assumption All clients are initialized with the same pre-trained model M0.
    Stated in Phase 1. This is a reasonable starting point but is an assumption about model initialization in the DFL setting.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Loss-Guided Model Sharing and Local Learning Correction in Decentralized Federated Learning for Crop Disease Classification." pith.science (2026). https://pith.science/paper/K6PAAW4U

@misc{pith2026250523063,
  author       = {Pith},
  title        = {Pith review of: Loss-Guided Model Sharing and Local Learning Correction in Decentralized Federated Learning for Crop Disease Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/K6PAAW4U}},
  note         = {Machine review of arXiv:2505.23063}
}
read the original abstract

Crop disease detection and classification is a critical challenge in agriculture, with major implications for productivity, food security, and environmental sustainability. While deep learning models such as CNN and ViT have shown excellent performance in classifying plant diseases from images, their large-scale deployment is often limited by data privacy concerns. Federated Learning (FL) addresses this issue, but centralized FL remains vulnerable to single-point failures and scalability limits. In this paper, we introduce a novel Decentralized Federated Learning (DFL) framework that uses validation loss (Loss_val) both to guide model sharing between peers and to correct local training via an adaptive loss function controlled by weighting parameter. We conduct extensive experiments using PlantVillage datasets with three deep learning architectures (ResNet50, VGG16, and ViT_B16), analyzing the impact of weighting parameter, the number of shared models, the number of clients, and the use of Loss_val versus Loss_train of other clients. Results demonstrate that our DFL approach not only improves accuracy and convergence speed, but also ensures better generalization and robustness across heterogeneous data environments making it particularly well-suited for privacy-preserving agricultural applications.

Figures

Figures reproduced from arXiv: 2505.23063 by the authors.

Figure 1
Figure 1. Architecture of the proposed approach. 4) Models aggregation : Let θ t k be the model parameters of each participant k at cycle t. The aim of this phase is to allow, for each participant k and at each cycle t, to locally aggregate the customer’s private model as well as the n best models received from Nk neighbors using the FedAvg algorithm. This algorithm aggregates the local models by averaging the weights to obta… view at source ↗
Figure 2
Figure 2. Example of images from our datasets. Table II EXPERIMENTAL PARAMETERS. Client Epoch/Client Batch Size Rounds Classes Learning rate Values 6, 18 1 16 50 4 0.01 experiments: grapes, apples and corn. These data are organized into different categories of healthy and diseased leaves. Table I provides a detailed description of these categories [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Impact of the λ parameter, in 6-client configuration, one best shared model, with λ = 0.25. Table IV RESULTS FROM THE STUDY OF THE NUMBER OF SHARED MODELS, IN 6-CLIENT CONFIGURATION, FIVE BEST SHARED MODELS. λ ResNet50 VGG16 ViT B16 F1-Score Accuracy F1-Score F1-Score Accuracy F1-Score 0.00 70,37 ± 10,09 75,78 ± 9,87 99,23 ± 0,49 99,23 ± 0,49 97,52 ± 1,40 97,55 ± 1,36 0.25 96,88 ± 4,75 96,97 ± 4,61 99,33 ± 0,68 99,3… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Impact of the number of shared models, in 6-client configuration, five best shared models, with [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Impact of using Lossval vs Losstrain in a 18-client configuration, five best shared models. other hand, in the case of VGG16 and ViT B16, the use of Losstrain generates a slight drop in performance for λ = 0.75, with minimal difference from the scenario using Lossval, …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

61 extracted references · 39 canonical work pages

  1. [49]

    Benchmarking federated strategies in peer-to-peer federated learning for biomedical data,

    J. L. Salmeron, I. Ar ´evalo, and A. Ruiz-Celma, “Benchmarking federated strategies in peer-to-peer federated learning for biomedical data,” Heliyon, vol. 9, no. 6, p. e16925, Jun. 2023. [Online]. Available: http://dx.doi.org/10.1016/j.heliyon.2023.e16925

  2. [28]

    Image- based crop disease detection with federated learning,

    D. Mamba Kabala, A. Hafiane, L. Bobelin, and R. Canals, “Image- based crop disease detection with federated learning,” Scientific Reports, vol. 13, no. 1, Nov. 2023. [Online]. Available: http: //dx.doi.org/10.1038/s41598-023-46218-5

  3. [56]

    Identification of plant leaf diseases using a nine-layer deep convolutional neural network,

    G. G. and A. P. J., “Identification of plant leaf diseases using a nine-layer deep convolutional neural network,” Computers and Electrical Engineering , vol. 76, p. 323–338, Jun. 2019. [Online]. Available: http://dx.doi.org/10.1016/j.compeleceng.2019.04.011

  4. [57]

    Identification of plant leaf diseases by deep learning based on channel attention and channel pruning,

    R. Chen, H. Qi, Y . Liang, and M. Yang, “Identification of plant leaf diseases by deep learning based on channel attention and channel pruning,” Frontiers in Plant Science , vol. 13, Nov. 2022. [Online]. Available: http://dx.doi.org/10.3389/fpls.2022.1023515

  5. [1]

    Rome: FAO, 2023

    Food and Agriculture Organization of the United Nations, World Food and Agriculture – Statistical Yearbook 2023 . Rome: FAO, 2023. [Online]. Available: https://doi.org/10.4060/cc7724en

  6. [2]

    Plant disease dynamics in a changing climate: impacts, molecular mechanisms, and climate-informed strategies for sustainable management,

    M. M. Hossain, F. Sultana, M. Mostafa, H. Ferdus, M. Rahman, J. A. Rana, S. S. Islam, S. Adhikary, A. Sannal, M. Al Emran Hosen, J. Nayeema, N. J. Emu, M. Kundu, S. K. Biswas, L. Farzana, and M. A. Al Sabbir, “Plant disease dynamics in a changing climate: impacts, molecular mechanisms, and climate-informed strategies for sustainable management,” Discover ...

  7. [3]

    Impact of pesticides use in agriculture: their benefits and hazards,

    W. Aktar, D. Sengupta, and A. Chowdhury, “Impact of pesticides use in agriculture: their benefits and hazards,” Interdisciplinary Toxicology, vol. 2, no. 1, pp. 1–12, Mar. 2009. [Online]. Available: https://doi.org/10.2478/v10102-009-0001-7

  8. [4]

    The global burden of pathogens and pests on major food crops,

    S. Savary, L. Willocquet, S. J. Pethybridge, P. Esker, N. McRoberts, and A. Nelson, “The global burden of pathogens and pests on major food crops,” Nature Ecology & amp; Evolution, vol. 3, no. 3, p. 430–439, Feb

Show all 61 references
  1. [5]

    Plant leaf disease detection, classification, and diagnosis using computer vision and artificial intelligence: A review,

    A. Bhargava, A. Shukla, O. P. Goswami, M. H. Alsharif, P. Uthansakul, and M. Uthansakul, “Plant leaf disease detection, classification, and diagnosis using computer vision and artificial intelligence: A review,” IEEE Access , vol. 12, p. 37443–37469, 2024. [Online]. Available:...

  2. [6]

    Feature engineering to identify plant diseases using image processing and artificial intelligence: A comprehensive review,

    S. M. Javidan, A. Banakar, K. Rahnama, K. A. Vakilian, and Y . Ampatzidis, “Feature engineering to identify plant diseases using image processing and artificial intelligence: A comprehensive review,” Smart Agricultural Technology, vol. 8, p. 100480, Aug. 2024. [Online]. Availa...

  3. [7]

    An ensemble of deep learning architectures for accurate plant disease classification,

    A. H. Ali, A. Youssef, M. Abdelal, and M. A. Raja, “An ensemble of deep learning architectures for accurate plant disease classification,” Ecological Informatics , vol. 81, p. 102618, Jul. 2024. [Online]. Available: http://dx.doi.org/10.1016/j.ecoinf.2024.102618 15

  4. [8]

    Machine learning- based detection and extraction of crop diseases: A comprehensive study on disease patterns for precision agriculture,

    P. Chaudhari, R. V . Patil, and P. N. Mahalle, “Machine learning- based detection and extraction of crop diseases: A comprehensive study on disease patterns for precision agriculture,” International Journal of Intelligent Systems and Applications in Engineering , vol. 12, no. ...

  5. [9]

    A review on machine learning and deep learning image-based plant disease classification for industrial farming systems,

    P. Sajitha, A. D. Andrushia, N. Anand, and M. Naser, “A review on machine learning and deep learning image-based plant disease classification for industrial farming systems,” Journal of Industrial Information Integration , vol. 38, p. 100572, Mar. 2024. [Online]. Available: ht...

  6. [10]

    Plant disease detection and classification by deep learning : A review,

    L. Li, S. Zhang, and B. Wang, “Plant disease detection and classification by deep learning : A review,” IEEE Access , vol. 9, p. 56683–56698,

  7. [11]

    Early disease detection in plants using cnn,

    Tejaswini, P. Rastogi, S. Dua, Manikanta, and V . Dagar, “Early disease detection in plants using cnn,” Procedia Computer Science , vol. 235, p. 3468–3478, 2024. [Online]. Available: http://dx.doi.org/10.1016/j.p rocs.2024.04.327

  8. [12]

    A review of deep learning techniques used in agriculture,

    I. Attri, L. K. Awasthi, T. P. Sharma, and P. Rathee, “A review of deep learning techniques used in agriculture,” Ecological Informatics, vol. 77, p. 102217, Nov. 2023. [Online]. Available: http://dx.doi.org/10.1016/j.ecoinf.2023.102217

  9. [13]

    Convolutional neural networks in detection of plant leaf diseases: A review,

    B. Tugrul, E. Elfatimi, and R. Eryigit, “Convolutional neural networks in detection of plant leaf diseases: A review,” Agriculture, vol. 12, no. 8, p. 1192, Aug. 2022. [Online]. Available: http: //dx.doi.org/10.3390/agriculture12081192

  10. [14]

    Identification of plant-leaf diseases using cnn and transfer-learning approach,

    S. M. Hassan, A. K. Maji, M. Jasi ´nski, Z. Leonowicz, and E. Jasi ´nska, “Identification of plant-leaf diseases using cnn and transfer-learning approach,” Electronics, vol. 10, no. 12, p. 1388, Jun. 2021. [Online]. Available: http://dx.doi.org/10.3390/electronics10121388

  11. [15]

    Effective plant disease diagnosis using vision transformer trained with leafy-generative adversarial network-generated images,

    A. K. Singh, A. Rao, P. Chattopadhyay, R. Maurya, and L. Singh, “Effective plant disease diagnosis using vision transformer trained with leafy-generative adversarial network-generated images,” Expert Systems with Applications, vol. 254, p. 124387, Nov. 2024. [Online]. Availabl...

  12. [16]

    A deep learning based approach for automated plant disease classification using vision transformer,

    Y . Borhani, J. Khoramdel, and E. Najafi, “A deep learning based approach for automated plant disease classification using vision transformer,” Scientific Reports , vol. 12, no. 1, Jul. 2022. [Online]. Available: http://dx.doi.org/10.1038/s41598-022-15163-0

  13. [17]

    A novel hierarchical framework for plant leaf disease detection using residual vision transformer,

    S. Vallabhajosyula, V . Sistla, and V . K. K. Kolli, “A novel hierarchical framework for plant leaf disease detection using residual vision transformer,” Heliyon, vol. 10, no. 9, p. e29912, May 2024. [Online]. Available: http://dx.doi.org/10.1016/j.heliyon.2024.e29912

  14. [18]

    A novel plant type, leaf disease and severity identification framework using cnn and transformer with multi-label method,

    B. Yang, M. Li, F. Li, Y . Wang, Q. Liang, R. Zhao, C. Li, and J. Wang, “A novel plant type, leaf disease and severity identification framework using cnn and transformer with multi-label method,” Scientific Reports , vol. 14, no. 1, May 2024. [Online]. Available: http://dx.doi...

  15. [19]

    Crop leaf disease recognition based on self- attention convolutional neural network,

    W. Zeng and M. Li, “Crop leaf disease recognition based on self- attention convolutional neural network,” Computers and Electronics in Agriculture , vol. 172, p. 105341, May 2020. [Online]. Available: http://dx.doi.org/10.1016/j.compag.2020.105341

  16. [20]

    Plant disease detection and classification techniques: a comparative study of the performances,

    W. B. Demilie, “Plant disease detection and classification techniques: a comparative study of the performances,” Journal of Big Data , vol. 11, no. 1, Jan. 2024. [Online]. Available: http://dx.doi.org/10.1186/s4053 7-023-00863-9

  17. [21]

    Cross-comparative review of machine learning for plant disease detection: apple, cassava, cotton and potato plants,

    J. D. Omaye, E. Ogbuju, G. Ataguba, O. Jaiyeoba, J. Aneke, and F. Oladipo, “Cross-comparative review of machine learning for plant disease detection: apple, cassava, cotton and potato plants,” Artificial Intelligence in Agriculture , vol. 12, p. 127–151, Jun. 2024. [Online]. A...

  18. [22]

    Agricultural data privacy and federated learning: A review of challenges and opportunities,

    R. Dembani, I. Karvelas, N. A. Akbar, S. Rizou, D. Tegolo, and S. Fountas, “Agricultural data privacy and federated learning: A review of challenges and opportunities,” Computers and Electronics in Agriculture , vol. 232, p. 110048, May 2025. [Online]. Available: http://dx.doi...

  19. [23]

    A review of federated learning in agriculture,

    K. R. ˇZalik and M. ˇZalik, “A review of federated learning in agriculture,” Sensors, vol. 23, no. 23, p. 9566, Dec. 2023. [Online]. Available: http://dx.doi.org/10.3390/s23239566

  20. [24]

    Crop disease prediction using deep learning in a federated learning environment: Ensuring data privacy and agricultural sustainability,

    S. Behera, N. Padhy, R. Panigrahi, and S. Kumar Kuanar, “Crop disease prediction using deep learning in a federated learning environment: Ensuring data privacy and agricultural sustainability,” Procedia Computer Science , vol. 254, p. 137–146, 2025. [Online]. Available: http:/...

  21. [25]

    A survey on federated learning: challenges and applications,

    J. Wen, Z. Zhang, Y . Lan, Z. Cui, J. Cai, and W. Zhang, “A survey on federated learning: challenges and applications,” International Journal of Machine Learning and Cybernetics , vol. 14, no. 2, p. 513–535, Nov

  22. [26]

    A survey on federated learning,

    C. Zhang, Y . Xie, H. Bai, B. Yu, W. Li, and Y . Gao, “A survey on federated learning,” Knowledge-Based Systems , vol. 216, p. 106775, Mar. 2021. [Online]. Available: http://dx.doi.org/10.1016/j.knosys.2021 .106775

  23. [27]

    A systematic review of federated learning: Challenges, aggregation methods, and development tools,

    B. S. Guendouzi, S. Ouchani, H. EL Assaad, and M. EL Zaher, “A systematic review of federated learning: Challenges, aggregation methods, and development tools,” Journal of Network and Computer Applications, vol. 220, p. 103714, Nov. 2023. [Online]. Available: http://dx.doi.org...

  24. [29]

    A survey on security and privacy of federated learning,

    V . Mothukuri, R. M. Parizi, S. Pouriyeh, Y . Huang, A. Dehghantanha, and G. Srivastava, “A survey on security and privacy of federated learning,” Future Generation Computer Systems , vol. 115, p. 619–640, Feb. 2021. [Online]. Available: http://dx.doi.org/10.1016/j.future.2020.1 0.007

  25. [30]

    Federated machine learning: Concept and applications,

    T. C. Qiang Yang, Yang Liu and Y . Tong, “Federated machine learning: Concept and applications,” 2019

  26. [31]

    Defta: A plug-and-play peer-to-peer decentralized federated learning framework,

    Y . Zhou, M. Shi, Y . Tian, Q. Ye, and J. Lv, “Defta: A plug-and-play peer-to-peer decentralized federated learning framework,” Information Sciences, vol. 670, p. 120582, Jun. 2024. [Online]. Available: http://dx.doi.org/10.1016/j.ins.2024.120582

  27. [32]

    Towards peer-to-peer federated learning: Algorithms and comparisons to centralized federated learning,

    D. M ¨aenp¨a¨a, “Towards peer-to-peer federated learning: Algorithms and comparisons to centralized federated learning,” Master’s thesis, Link¨oping University, Department of Computer and Information Sci- ence, 2021

  28. [33]

    Sparsfa: Towards robust and communication-efficient peer-to-peer federated learning,

    H. Wang, L. Mu ˜noz-Gonz´alez, M. Z. Hameed, D. Eklund, and S. Raza, “Sparsfa: Towards robust and communication-efficient peer-to-peer federated learning,” Computers & Security , vol. 129, p. 103182, Jun

  29. [34]

    Decentralized federated learning: A survey and perspective,

    L. Yuan, Z. Wang, L. Sun, P. S. Yu, and C. G. Brinton, “Decentralized federated learning: A survey and perspective,” 2023. [Online]. Available: https://arxiv.org/abs/2306.01603

  30. [35]

    A decentralized federated learning-based cancer survival prediction method with privacy protection,

    H. Chai, Y . Huang, L. Xu, X. Song, M. He, and Q. Wang, “A decentralized federated learning-based cancer survival prediction method with privacy protection,” Heliyon, vol. 10, no. 11, p. e31873, Jun. 2024. [Online]. Available: http://dx.doi.org/10.1016/j.heliyon.2024.e31873

  31. [36]

    Plantvillage dataset,

    A. Ali, “Plantvillage dataset,” 2015. [Online]. Available: https: //www.kaggle.com/datasets/abdallahalidev/plantvillage-dataset

  32. [37]

    Rice leaf disease identification and classification using machine learning techniques: A comprehensive review,

    R. Mukherjee, A. Ghosh, C. Chakraborty, J. N. De, and D. P. Mishra, “Rice leaf disease identification and classification using machine learning techniques: A comprehensive review,” Engineering Applications of Artificial Intelligence , vol. 139, p. 109639, Jan. 2025. [Online]. ...

  33. [38]

    A real time monitoring system for accurate plant leaves disease detection using deep learning,

    K. N. Rahman, S. C. Banik, R. Islam, and A. A. Fahim, “A real time monitoring system for accurate plant leaves disease detection using deep learning,” Crop Design , vol. 4, no. 1, p. 100092, Feb. 2025. [Online]. Available: http://dx.doi.org/10.1016/j.cropd.2024.100092

  34. [39]

    Federated learning: Crop classification in a smart farm decentralised network,

    G. Idoje, T. Dagiuklas, and M. Iqbal, “Federated learning: Crop classification in a smart farm decentralised network,” Smart Agricultural Technology, vol. 5, p. 100277, Oct. 2023. [Online]. Available: http://dx.doi.org/10.1016/j.atech.2023.100277

  35. [40]

    A comprehensive review on federated learning based models for healthcare applications,

    S. Sharma and K. Guleria, “A comprehensive review on federated learning based models for healthcare applications,” Artificial Intelligence in Medicine , vol. 146, p. 102691, Dec. 2023. [Online]. Available: http://dx.doi.org/10.1016/j.artmed.2023.102691

  36. [41]

    A federated learning approach for smart healthcare systems,

    A. Mishra, S. Saha, S. Mishra, and P. Bagade, “A federated learning approach for smart healthcare systems,” CSI Transactions on ICT , vol. 11, no. 1, p. 39–44, Apr. 2023. [Online]. Available: http://dx.doi.org/10.1007/s40012-023-00382-1

  37. [42]

    Federated learning for medical image analysis: A survey,

    H. Guan, P.-T. Yap, A. Bozoki, and M. Liu, “Federated learning for medical image analysis: A survey,” Pattern Recognition, vol. 151, p. 110424, Jul. 2024. [Online]. Available: http://dx.doi.org/10.1016/j.pat cog.2024.110424

  38. [43]

    A review on federated learning in computational pathology,

    L. A. Schoenpflug, Y . Nie, F. Sheikhzadeh, and V . H. Koelzer, “A review on federated learning in computational pathology,” Computational and Structural Biotechnology Journal , vol. 23, p. 3938–3945, Dec. 2024. [Online]. Available: http://dx.doi.org/10.1016/j.csbj.2024.10.037

  39. [44]

    A contemporary survey of recent advances in federated learning: Taxonomies, applications, and challenges,

    M. H. Alsharif, R. Kannadasan, W. Wei, K. S. Nisar, and A.- H. Abdel-Aty, “A contemporary survey of recent advances in federated learning: Taxonomies, applications, and challenges,” Internet 16 of Things , vol. 27, p. 101251, Oct. 2024. [Online]. Available: http://dx.doi.org/1...

  40. [45]

    A review of federated learning applications in intrusion detection systems,

    A. Belenguer, J. A. Pascual, and J. Navaridas, “A review of federated learning applications in intrusion detection systems,” Computer Networks, vol. 258, p. 111023, Feb. 2025. [Online]. Available: http://dx.doi.org/10.1016/j.comnet.2024.111023

  41. [46]

    Federated learning for solar energy applications: A case study on real-time fault detection,

    I. A. Abdelmoula, H. Oufettoul, N. Lamrini, S. Motahhir, A. Mehdary, and M. E. Aroussi, “Federated learning for solar energy applications: A case study on real-time fault detection,” Solar Energy , vol. 282, p. 112942, Nov. 2024. [Online]. Available: http://dx.doi.org/10.1016/...

  42. [47]

    Efficient training: Federated learning cost analysis,

    R. Teixeira, L. Almeida, M. Antunes, D. Gomes, and R. L. Aguiar, “Efficient training: Federated learning cost analysis,” Big Data Research , vol. 40, p. 100510, May 2025. [Online]. Available: http://dx.doi.org/10.1016/j.bdr.2025.100510

  43. [48]

    Federated learning: Centralized and p2p for a siamese deep learning model for diabetes foot ulcer classification,

    M. S. A. Toofanee, M. Hamroun, S. Dowlut, K. Tamine, V . Petit, A. K. Duong, and D. Sauveron, “Federated learning: Centralized and p2p for a siamese deep learning model for diabetes foot ulcer classification,” Applied Sciences , vol. 13, no. 23, p. 12776, Nov. 2023. [Online]. ...

  44. [50]

    Decentralized federated learning based on blockchain: concepts, framework, and challenges,

    H. Zhang, S. Jiang, and S. Xuan, “Decentralized federated learning based on blockchain: concepts, framework, and challenges,” Computer Communications, vol. 216, p. 140–150, Feb. 2024. [Online]. Available: http://dx.doi.org/10.1016/j.comcom.2023.12.042

  45. [51]

    Consensus-driven hyperparameter optimization for accelerated model convergence in decentralized federated learning,

    A. N. Khan, Q. W. Khan, A. Rizwan, R. Ahmad, and D. H. Kim, “Consensus-driven hyperparameter optimization for accelerated model convergence in decentralized federated learning,” Internet of Things , vol. 30, p. 101476, Mar. 2025. [Online]. Available: http://dx.doi.org/10.1016/...

  46. [52]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” 2015

  47. [53]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” 2015

  48. [54]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Transformers for image recognition at scale,” 2021

  49. [55]

    A systematic analysis of performance measures for classification tasks,

    M. Sokolova and G. Lapalme, “A systematic analysis of performance measures for classification tasks,” Information Processing & Management, vol. 45, no. 4, pp. 427–437, Jul. 2009. [Online]. Available: https://doi.org/10.1016/j.ipm.2009.03.002

  50. [2019]

    Available: http://dx.doi.org/10.1038/s41559-018-0793-y

    [Online]. Available: http://dx.doi.org/10.1038/s41559-018-0793-y

  51. [2021]

    Available: http://dx.doi.org/10.1109/access.2021.30696 46

    [Online]. Available: http://dx.doi.org/10.1109/access.2021.30696 46

  52. [2022]

    Available: http://dx.doi.org/10.1007/s13042-022-01647 -y

    [Online]. Available: http://dx.doi.org/10.1007/s13042-022-01647 -y

  53. [2023]

    Available: http://dx.doi.org/10.1016/j.cose.2023.103182

    [Online]. Available: http://dx.doi.org/10.1016/j.cose.2023.103182

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.