Pith. sign in

REVIEW 3 major objections 5 minor 45 references

Sparse Self-Federated Learning for Energy Efficient Cooperative Intelligence in Society 5.0

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper argues that pruning and quantizing the exchanged models gives self-federated learning large energy and bandwidth savings without meaningful accuracy loss.

desk verdict A clear, honest roadmap for compressing proximity-based FL, but the central accuracy claim is unmeasured end-to-end. read the letter →

arxiv 2507.07613 v1 pith:ZFNDKUHP submitted 2025-07-10 cs.LG

Davide Domini , Laura Erhan , Gianluca Aguzzi , Lucia Cavallaro , Amirhossein Douzandeh Zenoozi , Antonio Liotta , Mirko Viroli This is my paper · ORCID
classification cs.LG
keywords GreenFederatedLearningSparseNeuralNetworksQuantizationSociety5.0Self-federatedAggregatecomputingEnergyefficiencyNon-IIDdata
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

The paper proposes SParSeFuL, a variant of Proximity-based Self-Federated Learning in which the models that devices exchange for similarity checks and aggregation are sparsified and quantized. The claim is that this compression preserves the ability of a loss-based similarity metric to group devices with similar data distributions, so federations still converge to the true regions of non-IID data, while cutting inference time and energy and at least partially reducing communication overhead. The authors support the claim with preliminary measurements: pruning 30% of weights cuts inference time by nearly two-thirds and energy by about half with no visible accuracy drop, and 8-bit quantization shrinks a model from 1.8 MB to 454 KB. The rest of the paper is a research roadmap for integrating compression more deeply, including adaptive sparsity and real-world deployment, rather than a full evaluation of the complete framework.

What carries the argument

The load-bearing mechanism is the loss-based similarity check performed on compressed models. In PSFL, two devices $s_i$ and $s_j$ exchange trained models, evaluate each other's model on their own local data, and form a federation when the combined loss $d_{s_{i,j}} = L_{i,j} + L_{j,i}$ falls below a threshold $\tau$; this proxy stands in for direct comparison of private datasets. SParSeFuL's compression strategy $C$ with sparsification ratio $\psi$ is applied before the model is broadcast, so the similarity metric and the eventual FedAvg aggregation operate on sparse, quantized models. The Self-Organizing Coordination Region pattern supplies the distributed leader election (S-block), gradient-based propagation (G-block), and collection/aggregation (C-block) that turn the pairwise similarity decisions into stable federations.

What would settle it

Run SParSeFuL on a non-IID benchmark with, say, four known data subregions, sweeping the sparsification ratio $\psi$; if the number of converged federations diverges from four, or accuracy drops beyond the dense baseline's variance, at $\psi$ values below 0.8, then the central claim that compression preserves federation formation is falsified. The paper itself warns that $\psi \geq 0.8$ can break differentiation.

Watch

Extended reading notes

Core claim

The central discovery, on the paper's own terms, is that the resource bottleneck of self-federated learning can be attacked at the model-exchange level without sacrificing federation quality. SParSeFuL extends PSFL, where devices elect leaders through an aggregate-computing Self-Organizing Coordination Region pattern and form federations by pairwise loss-based similarity $d_{s_{i,j}} = L_{i,j} + L_{j,i}$; the extension compresses each model with a compression strategy $C$ at sparsity ratio $\psi$ before local training, broadcast, similarity evaluation, and aggregation. The authors report that aggressive sparsification is tolerable up to a point: measured on an MLP trained on Extended MNIST, $\psi = 0.3$ reduces inference time by nearly two-thirds and relative energy consumption to about 41% of the dense baseline with no noticeable accuracy loss, while performance degrades only near $\psi = 0.9$. Quantization at 8 bits reduces the exchanged model to about a quarter of its original size. The intended upshot is that a partially compressed pipeline can run on resource-constrained IoT devices and still converge to the correct number of federations.

Load-bearing premise

The similarity metric computed on compressed models must remain a reliable proxy for true data-distribution similarity at the sparsity levels that deliver the reported savings.

Editorial extensions

If this is right

  • If the claim holds, a moderately sparse model exchange near $\psi = 0.3$ can be used in the similarity check without degrading federation formation, because the loss proxy remains informative at that sparsity.
  • Energy per device falls roughly in half at $\psi = 0.3$, so the aggregate energy budget of large IoT deployments would shrink correspondingly.
  • Eight-bit quantization cuts model size to about one fourth, which directly reduces the bandwidth of every model broadcast and collection in the federated loop.
  • Federations should still converge to the true number of data distributions, because the similarity check is evaluated after compression, not on full models.
  • Extreme sparsification near $\psi \geq 0.8$ is expected to break distribution differentiation and must be avoided.

Reading between the lines

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

  • A testable extension the paper leaves implicit is to make the sparsity ratio adaptive per device, using the converged federation count as feedback to lower compression when federations fail to stabilize.
  • Because iterative pruning can leave devices with architecturally diverse sparse models, the FedAvg step may need a mask-alignment or topology-matching procedure; comparing plain FedAvg against such a procedure on sparse models would be a concrete next experiment.
  • The reported energy figures measure single-model inference, not the full federated loop, so an honest extrapolation is that end-to-end measurements are needed to see whether communication energy, rather than compute energy, dominates the savings in bandwidth-limited IoT settings.
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 / 5 minor

Summary. The paper proposes SParSeFuL, a sparse, quantized extension of Proximity-based Self-Federated Learning (PSFL) for large-scale IoT systems. The core mechanism combines aggregate-computing-based self-organization (SCR pattern) with neural network compression: models are sparsified and quantized before the loss-based similarity check that determines federation membership, and compressed models are then used for local training, exchange, and aggregation. The contributions are stated as (i) integrating compression into PSFL, (ii) a research roadmap for implementation and evaluation, and (iii) a discussion of impacts and challenges. The only quantitative evidence is a standalone MLP experiment on EMNIST showing reduced inference time and energy after post-hoc sparsification, plus a report that 8-bit quantization reduces model size from 1.8MB to 454KB. The abstract and conclusion claim that SParSeFuL reduces energy and communication overhead while maintaining competitive accuracy even under significant sparsification.

Significance. If established, the central claim would be relevant to green federated learning in non-IID IoT settings: combining self-organizing federation formation with compression addresses a real gap, since most prior work handles non-IID data, compression, or decentralization separately but not jointly. The paper is clearly written and provides a concrete algorithmic skeleton (Algorithm 1), a sensible roadmap, and honest preliminary measurements using a public dataset and the PyJoules energy tool. However, as it stands, the paper is a position/roadmap contribution rather than a validated method: no experiment exercises Algorithm 1 end-to-end, and the one load-bearing component that is new relative to PSFL—compressed-model loss similarity for federation formation—is explicitly conceded in Section III-B to fail at high sparsity and is never measured. The claimed energy, bandwidth, and accuracy benefits of SParSeFuL therefore remain unsubstantiated.

major comments (3)
  1. [Section V, Table I, Figure 4] The only quantitative evidence for the central claim is a post-pruned, standalone MLP: inference time, accuracy, and energy are measured outside any federated loop. There is no end-to-end run of Algorithm 1, no measurement of federation formation quality (e.g., whether federations converge to the true subregions), no federated accuracy after sparsification/quantization, and no accounting of communication or aggregation energy/bandwidth for SParSeFuL versus the uncompressed PSFL baseline. The abstract and conclusion claim that SParSeFuL 'can reduce power consumption and (partially) communication overhead while maintaining competitive accuracy,' but these quantities are never measured on SParSeFuL. The energy numbers in Table I are also unexplained in their steep non-monotonic drop from 41% at psi=0.3 to 3% at psi=0.5, and no variance or repetition information is given.
  2. [Section III-B and Algorithm 1, lines 3-5] The load-bearing premise is that the loss-based similarity metric d_{si,j} computed on compressed models remains a faithful proxy for data-distribution similarity. Algorithm 1 compresses the model before the similarity check and before local training, so the fidelity of d_{si,j} under sparsification and quantization determines whether federations form correctly. The paper itself states that at an 'excessively high sparsity level (e.g., psi >= 0.8)' the similarity check may fail to differentiate distributions, leading to inaccurate federation formation. Yet Figure 4 reports large inference savings precisely at psi in {0.7, 0.9}, and no experiment determines the threshold at which compression breaks federation formation. Without this measurement, the conclusion that SParSeFuL maintains accuracy 'even with significant sparsification' is unsupported.
  3. [Section IV-A] The research roadmap describes the integration of compression into the similarity check as a 'crucial first step' that must be verified before compression is extended to the learning phase: 'Once it is verified that this approximation enables effective federation formation without degrading accuracy...'. This is an explicit statement that the core mechanism of SParSeFuL has not yet been validated. The manuscript therefore overstates its contribution when it says in the abstract that SParSeFuL 'bridges this gap' and in the conclusion that it 'can reduce power consumption... while maintaining competitive accuracy.' The available evidence is consistent with the paper being a proposal, not with the paper having demonstrated a working method.
minor comments (5)
  1. [Section III-B] The phrase 'the similarity is computed as follow' should read 'as follows'; the enumerated steps (Models Exchange, Cross Validation, Similarity Metric Computation, Federation Decision) would be clearer as a numbered list with mathematical notation for L_{i,j} and d_{si,j} defined in one place.
  2. [Algorithm 1] Line 5 broadcasts M^{t,c}_i, the compressed model after local training, but the relationship between the broadcast model and the model used in the similarity metric computation (which evaluates received models on local data) is not fully specified; in particular, it is ambiguous whether the uncompressed local model is retained for training while only the exchanged copy is compressed, or whether compression affects the locally trained model itself. Clarifying this would make the algorithm reproducible.
  3. [Section V] The text says the experiment used an 'MLP network trained on the Extended MNIST dataset,' but Section IV-B mentions MNIST and CIFAR-10 as datasets for future evaluation; stating explicitly which dataset (EMNIST versus MNIST) was used in the reported experiment would avoid confusion.
  4. [Section V, Table I] The relative energy values would benefit from error bars or multiple runs, and the sudden drop between psi=0.3 and psi=0.5 should be explained (e.g., hardware caching, threshold effects in the pruning implementation, or measurement granularity).
  5. [Section II-C] The description of SCR relies heavily on the cited reference for the meaning of the S-block, G-block, and C-block; a one-sentence definition of each block at first use would make Section III more self-contained.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the compression gains are independently measured on a standalone model, and the PSFL baseline is prior work, not a fitted input.

full rationale

SParSeFuL's efficiency claims rest on direct measurements, not on derivation from its own assumptions. Figure 4 and Table I report inference time, accuracy, and PyJoules-measured energy for a post-pruned MLP on EMNIST, and the quantization result is given as a concrete model-size reduction at 8-bit integer precision. None of these quantities is fitted, renamed, or derived from the cited PSFL framework; PSFL from [8] is the algorithmic substrate, not the source of the efficiency numbers. The framework's core mechanism, the loss-based similarity metric, is inherited from PSFL and is admittedly fragile: Section III-B explicitly states that at high sparsity (psi >= 0.8) the process may fail to differentiate distributions, and Section IV-A lists verifying that compressed similarity enables effective federation formation as a future objective. These are empirical validity gaps and correctness risks, not circular reductions. The paper also cites several prior works by overlapping authors, including PSFL [8], FBFL [36], and ProFed [43], but these citations provide context and baselines; the central claim about compression does not reduce to any of them by construction. No equation in the paper is equivalent to another by definition, and no fitted parameter is relabeled as a prediction. Therefore the honest finding is no significant circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The proposal rests on the correctness of the PSFL proximity assumption, on the loss-based similarity proxy surviving compression, and on the SCR pattern's self-stabilization guarantees, all taken from prior work without demonstration here. The energy and accuracy claims depend on user-chosen hyperparameters psi and tau, which are not specified for the federated setting.

free parameters (3)
  • sparsification ratio psi = tested 0.0, 0.3, 0.5, 0.7, 0.9; no principled selection
    Controls the tradeoff between resource savings and accuracy; the paper's claims depend on choosing psi where accuracy is preserved, e.g., 0.3.
  • similarity threshold tau = not specified
    Determines federation formation; no guidance on setting tau is given.
  • compression strategy C = post-pruning with unspecified criterion; 8-bit integer quantization
    The choice of compression strategy affects both similarity checks and model aggregation; the paper does not specify or compare strategies.
assumptions (4)
  • domain assumption Devices in spatial proximity experience similar data distributions
    Assumed in Section II-A and III-A to justify proximity-based federation; cited from prior work [19], [20].
  • domain assumption Cross-validation loss on compressed models is a reliable proxy for data distribution similarity
    Section III-B defines similarity as the sum of cross losses; the paper notes high sparsity (psi >= 0.8) can break this, so the assumption is load-bearing and partially acknowledged.
  • domain assumption The SCR pattern (S-block, G-block, C-block) provides correct self-stabilizing federation formation
    Relied on from prior aggregate computing literature [35], [36] without proof in this paper.
  • domain assumption FedAvg aggregation works for models with differing sparsity patterns
    Section V admits iterative pruning can create architecturally diverse sparse models, making aggregation complex; the paper does not solve this.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sparse Self-Federated Learning for Energy Efficient Cooperative Intelligence in Society 5.0." pith.science (2026). https://pith.science/paper/ZFNDKUHP

@misc{pith2026250707613,
  author       = {Pith},
  title        = {Pith review of: Sparse Self-Federated Learning for Energy Efficient Cooperative Intelligence in Society 5.0},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZFNDKUHP}},
  note         = {Machine review of arXiv:2507.07613}
}
read the original abstract

Federated Learning offers privacy-preserving collaborative intelligence but struggles to meet the sustainability demands of emerging IoT ecosystems necessary for Society 5.0-a human-centered technological future balancing social advancement with environmental responsibility. The excessive communication bandwidth and computational resources required by traditional FL approaches make them environmentally unsustainable at scale, creating a fundamental conflict with green AI principles as billions of resource-constrained devices attempt to participate. To this end, we introduce Sparse Proximity-based Self-Federated Learning (SParSeFuL), a resource-aware approach that bridges this gap by combining aggregate computing for self-organization with neural network sparsification to reduce energy and bandwidth consumption.

Figures

Figures reproduced from arXiv: 2507.07613 by the authors.

Figure 1
Figure 1. Visual representation of data skewness between differ [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed algorithm, with highlighted components where compression techniques discussed in Sec [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Graphical representation of federations evolution over time. As in Figure 1, we have four different subregions ( [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Impact of sparsification level ψ on neural network performance. TABLE I: Relative energy consumption of sparse neural networks at different sparsification levels (ψ), normalized with respect to the dense baseline (ψ = 0.0). Energy consumption has been measured using Py…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 39 canonical work pages

  1. [1]

    What is society 5.0,

    A. Deguchi, C. Hirai, H. Matsuoka, T. Nakano, K. Oshima, M. Tai, and S. Tani, “What is society 5.0,” Society, vol. 5, no. 0, pp. 1–24, 2020

    work page 2020

  2. [2]

    Foundations of cooperative AI,

    V . Conitzer and C. Oesterheld, “Foundations of cooperative AI,” in Thirty-Seventh AAAI Conference on Artificial Intelligence, AAAI 2023, Thirty-Fifth Conference on Innovative Applications of Artificial Intelligence, IAAI 2023, Thirteenth Symposium on Educational Advances in Artificial Intelligence, EAAI 2023, Washington, DC, USA, February 7-14, 2023 , B....

    doi:10.1609/aaai.v37i13.26791 2023

  3. [3]

    An efficient framework for clustered federated learning,

    A. Ghosh, J. Chung, D. Yin, and K. Ramchandran, “An efficient framework for clustered federated learning,” IEEE Trans. Inf. Theory, vol. 68, no. 12, pp. 8076–8091, 2022. [Online]. Available: https://doi.org/10.1109/TIT.2022.3192506

    work page arXiv 2022

  4. [4]

    Towards effective clustered federated learning: A peer-to-peer framework with adaptive neighbor matching,

    Z. Li, J. Lu, S. Luo, D. Zhu, Y . Shao, Y . Li, Z. Zhang, Y . Wang, and C. Wu, “Towards effective clustered federated learning: A peer-to-peer framework with adaptive neighbor matching,” IEEE Trans. Big Data , vol. 10, no. 6, pp. 812–826, 2024. [Online]. Available: https://doi.org/10.1109/TBDATA.2022.3222971

    work page arXiv 2024

  5. [5]

    Iot device friendly and communication-efficient federated learning via joint model pruning and quantization,

    P. Prakash, J. Ding, R. Chen, X. Qin, M. Shu, Q. Cui, Y . Guo, and M. Pan, “Iot device friendly and communication-efficient federated learning via joint model pruning and quantization,” IEEE Internet Things J., vol. 9, no. 15, pp. 13 638–13 650, 2022. [Online]. Available: https://doi.org/10.1109/JIOT.2022.3145865

    work page arXiv 2022

  6. [6]

    Towards self-adaptive cooperative learning in collective systems,

    D. Domini, “Towards self-adaptive cooperative learning in collective systems,” in IEEE International Conference on Autonomic Computing and Self-Organizing Systems, ACSOS 2024 - Companion, Aarhus, Denmark, September 16-20, 2024 . IEEE, 2024, pp. 158–160. [Online]. Available: https://doi.org/10.1109/ACSOS-C63493.2024.00049

    work page arXiv 2024

  7. [7]

    Decentralized learning works: An empirical comparison of gossip learning and federated learning,

    I. Heged ¨us, G. Danner, and M. Jelasity, “Decentralized learning works: An empirical comparison of gossip learning and federated learning,” J. Parallel Distributed Comput. , vol. 148, pp. 109–124, 2021. [Online]. Available: https://doi.org/10.1016/j.jpdc.2020.10.006

    work page doi:10.1016/j.jpdc.2020.10.006 2021

  8. [8]

    Proximity-based self-federated learning,

    D. Domini, G. Aguzzi, N. Farabegoli, M. Viroli, and L. Esterle, “Proximity-based self-federated learning,” in IEEE International Conference on Autonomic Computing and Self- Organizing Systems, ACSOS 2024, Aarhus, Denmark, September 16-20, 2024 . IEEE, 2024, pp. 139–144. [Online]. Available: https://doi.org/10.1109/ACSOS61780.2024.00033

    work page arXiv 2024

Show all 45 references
  1. [9]

    Communication-efficient learning of deep networks from decentralized data,

    B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” in Proceedings of the 20th International Conference on Artificial Intelligence and Statistics, AISTATS 2017, 20-22 April 2017, Fort Laud...

    2017

  2. [10]

    Federated learning of deep networks using model averaging,

    H. B. McMahan, E. Moore, D. Ramage, and B. A. y Arcas, “Federated learning of deep networks using model averaging,” CoRR, vol. abs/1602.05629, 2016. [Online]. Available: http://arxiv.org/abs/1602.05629

    2016 arXiv

  3. [11]

    Personalized federated learning for cross-city traffic prediction,

    Y . Zhang, H. Lu, N. Liu, Y . Xu, Q. Li, and L. Cui, “Personalized federated learning for cross-city traffic prediction,” in Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, IJCAI 2024, Jeju, South Korea, August 3-9, 2024 . ijcai.org, ...

    2024

  4. [12]

    A survey on federated learning in intelligent transportation systems,

    R. Zhang, H. Wang, B. Li, X. Cheng, and L. Yang, “A survey on federated learning in intelligent transportation systems,” CoRR, vol. abs/2403.07444, 2024. [Online]. Available: https://doi.org/10.48550/arXiv.2403.07444

    2024 doi arXiv

  5. [13]

    A survey on federated learning for resource-constrained iot devices,

    A. Imteaj, U. Thakker, S. Wang, J. Li, and M. H. Amini, “A survey on federated learning for resource-constrained iot devices,” IEEE Internet Things J. , vol. 9, no. 1, pp. 1–24, 2022. [Online]. Available: https://doi.org/10.1109/JIOT.2021.3095077

    2022

  6. [14]

    Federated learning on non-iid data silos: An experimental study,

    Q. Li, Y . Diao, Q. Chen, and B. He, “Federated learning on non-iid data silos: An experimental study,” in Proceedings of the International Conference on Data Engineering . IEEE, 2022, pp. 965–978. [Online]. Available: https://doi.org/10.1109/ICDE53745.2022.00077

    2022

  7. [15]

    Imteaj, K

    A. Imteaj, K. Mamun Ahmed, U. Thakker, S. Wang, J. Li, and M. H. Amini, Federated Learning for Resource-Constrained IoT Devices: Panoramas and State of the Art . Cham: Springer International Publishing, 2023, pp. 7–27. [Online]. Available: https://doi.org/10.1007/978-3-031-11748-0 2

    2023 doi

  8. [16]

    Federated optimization in heterogeneous networks,

    T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V . Smith, “Federated optimization in heterogeneous networks,” in Proceedings of the Third Conference on Machine Learning and Systems, MLSys 2020, Austin, TX, USA, March 2-4, 2020 , I. S. Dhillon, D. S. Papailiopoulos...

    2020

  9. [17]

    SCAFFOLD: stochastic controlled averaging for federated learning,

    S. P. Karimireddy, S. Kale, M. Mohri, S. J. Reddi, S. U. Stich, and A. T. Suresh, “SCAFFOLD: stochastic controlled averaging for federated learning,” in Proceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event , ser. Proce...

    2020

  10. [18]

    Unsupervised graph structure-assisted personalized federated learning,

    X. Li, X. Chen, B. Tang, S. Wang, Y . Xuan, and Z. Zhao, “Unsupervised graph structure-assisted personalized federated learning,” in Proceedings of the European Conference on Artificial Intelligence , ser. Frontiers in Artificial Intelligence and Applications, vol. 372. IOS Pr...

    2023 doi

  11. [19]

    Deep learning in multiagent systems,

    L. Esterle, “Deep learning in multiagent systems,” in Deep Learning for Robot Perception and Cognition . Elsevier, 2022, pp. 435–460

    2022

  12. [20]

    Neighbor-based decentralized training strategies for multi-agent reinforcement learning,

    N. Malucelli, D. Domini, G. Aguzzi, and M. Viroli, “Neighbor-based decentralized training strategies for multi-agent reinforcement learning,” in Proceedings of the 40th ACM/SIGAPP Symposium on Applied Com- puting, SAC 2025, Catania, Italy, March 31-April 4, 2025. ACM, 2024, pp. 3–10

    2025

  13. [21]

    Field-based coordination for federated learning,

    D. Domini, G. Aguzzi, L. Esterle, and M. Viroli, “Field-based coordination for federated learning,” in Coordination Models and Languages - 26th IFIP WG 6.1 International Conference, COORDINATION 2024, Held as Part of the 19th International Federated Conference on Distributed C...

    2024

  14. [22]

    Edge AI: On-demand acceler- ating deep neural network inference via edge computing,

    E. Li, L. Zeng, Z. Zhou, and X. Chen, “Edge AI: On-demand acceler- ating deep neural network inference via edge computing,” IEEE Trans. Wirel. Commun., vol. 19, no. 1, pp. 447–457, Jan. 2020

    2020

  15. [23]

    Pruning and quantization for deep neural network acceleration: A survey,

    T. Liang, J. Glossner, L. Wang, S. Shi, and X. Zhang, “Pruning and quantization for deep neural network acceleration: A survey,” Neuro- computing, vol. 461, pp. 370–403, Oct. 2021

    2021

  16. [24]

    Comparing training of sparse to classic neural networks for binary classification in medical data,

    L. Erhan, A. Liotta, and L. Cavallaro, “Comparing training of sparse to classic neural networks for binary classification in medical data,” in Lecture Notes in Computer Science , ser. Lecture notes in computer science. Cham: Springer Nature Switzerland, 2025, pp. 101–106

    2025

  17. [25]

    Resprune: An energy-efficient restorative filter pruning method using stochastic optimization for accelerating cnn,

    A. Jayasimhan and P. P., “Resprune: An energy-efficient restorative filter pruning method using stochastic optimization for accelerating cnn,” Pattern Recognition, vol. 155, p. 110671, Nov 2024. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0031320324004229

    2024

  18. [26]

    Unveiling the power of sparse neural networks for feature selection,

    Z. Atashgahi, T. Liu, M. Pechenizkiy, R. Veldhuis, D. C. Mocanu, and M. van der Schaar, “Unveiling the power of sparse neural networks for feature selection,” Oct. 2024. [Online]. Available: https://arxiv.org/abs/2408.04583

    2024 arXiv

  19. [27]

    Quantizing deep convolutional networks for efficient inference: A whitepaper,

    R. Krishnamoorthi, “Quantizing deep convolutional networks for efficient inference: A whitepaper,” 2018. [Online]. Available: https://arxiv.org/abs/1806.08342

    2018 arXiv

  20. [28]

    Aggregate programming for the internet of things,

    J. Beal, D. Pianini, and M. Viroli, “Aggregate programming for the internet of things,” Computer, vol. 48, no. 9, pp. 22–30, 2015. [Online]. Available: https://doi.org/10.1109/MC.2015.261

    2015 doi

  21. [29]

    Macroprogramming: Concepts, state of the art, and opportunities of macroscopic behaviour modelling,

    R. Casadei, “Macroprogramming: Concepts, state of the art, and opportunities of macroscopic behaviour modelling,” ACM Comput. Surv., vol. 55, no. 13s, pp. 275:1–275:37, 2023. [Online]. Available: https://doi.org/10.1145/3579353

    2023 doi

  22. [30]

    Co-fields: A physically inspired approach to motion coordination,

    M. Mamei, F. Zambonelli, and L. Leonardi, “Co-fields: A physically inspired approach to motion coordination,” IEEE Pervasive Comput., vol. 3, no. 2, pp. 52–61, 2004. [Online]. Available: https://doi.org/10.1109/MPRV .2004.1316820

    2004 arXiv

  23. [31]

    A higher-order calculus of computational fields,

    G. Audrito, M. Viroli, F. Damiani, D. Pianini, and J. Beal, “A higher-order calculus of computational fields,” ACM Trans. Comput. Log., vol. 20, no. 1, pp. 5:1–5:55, 2019. [Online]. Available: https://doi.org/10.1145/3285956

    2019 doi

  24. [32]

    Space-time universality of field calculus,

    G. Audrito, J. Beal, F. Damiani, and M. Viroli, “Space-time universality of field calculus,” in Coordination Models and Languages - 20th IFIP WG 6.1 International Conference, COORDINATION 2018, Held as Part of the 13th International Federated Conference on Distributed Computin...

    2018

  25. [33]

    Engineering resilient collective adaptive systems by self-stabilisation,

    M. Viroli, G. Audrito, J. Beal, F. Damiani, and D. Pianini, “Engineering resilient collective adaptive systems by self-stabilisation,” ACM Trans. Model. Comput. Simul. , vol. 28, no. 2, pp. 16:1–16:28, 2018. [Online]. Available: https://doi.org/10.1145/3177774

    2018 doi

  26. [34]

    Self-adaptation to device distribution in the internet of things,

    J. Beal, M. Viroli, D. Pianini, and F. Damiani, “Self-adaptation to device distribution in the internet of things,” ACM Trans. Auton. Adapt. Syst., vol. 12, no. 3, pp. 12:1–12:29, 2017. [Online]. Available: https://doi.org/10.1145/3105758

    2017 doi

  27. [35]

    Self-organising coordination regions: A pattern for edge computing,

    R. Casadei, D. Pianini, M. Viroli, and A. Natali, “Self-organising coordination regions: A pattern for edge computing,” in Coordination Models and Languages - 21st IFIP WG 6.1 International Conference, COORDINATION 2019, Held as Part of the 14th International Federated Confere...

    2019

  28. [36]

    FBFL: A field-based coordination approach for data heterogeneity in federated learning,

    D. Domini, G. Aguzzi, L. Esterle, and M. Viroli, “FBFL: A field-based coordination approach for data heterogeneity in federated learning,” CoRR, vol. abs/2502.08577, 2025. [Online]. Available: https://doi.org/10.48550/arXiv.2502.08577

    2025 doi

  29. [37]

    Scarlib: Towards a hybrid toolchain for aggregate computing and many-agent reinforcement learning,

    D. Domini, F. Cavallari, G. Aguzzi, and M. Viroli, “Scarlib: Towards a hybrid toolchain for aggregate computing and many-agent reinforcement learning,” Sci. Comput. Program. , vol. 238, p. 103176, 2024. [Online]. Available: https://doi.org/10.1016/j.scico.2024.103176

    2024

  30. [38]

    Field-informed reinforcement learning of collective tasks with graph neural networks,

    G. Aguzzi, M. Viroli, and L. Esterle, “Field-informed reinforcement learning of collective tasks with graph neural networks,” in IEEE International Conference on Autonomic Computing and Self-Organizing Systems, ACSOS 2023, Toronto, ON, Canada, September 25-29, 2023 . IEEE, 202...

    2023

  31. [39]

    Scalable training of artificial neural networks with adaptive sparse connectivity inspired by network science,

    D. C. Mocanu, E. Mocanu, P. Stone, P. H. Nguyen, M. Gibescu, and A. Liotta, “Scalable training of artificial neural networks with adaptive sparse connectivity inspired by network science,” Nature communica- tions, vol. 9, no. 1, p. 2383, 2018

    2018

  32. [40]

    Mnist handwritten digit database,

    Y . LeCun, C. Cortes, C. Burges et al. , “Mnist handwritten digit database,” 2010

    2010

  33. [41]

    Cifar-10 (cana- dian institute for advanced research)

    A. Krizhevsky, V . Nair, and G. Hinton, “Cifar-10 (cana- dian institute for advanced research).” [Online]. Available: http://www.cs.toronto.edu/ kriz/cifar.html

  34. [42]

    Flower: A friendly federated learning research framework,

    D. J. Beutel, T. Topal, A. Mathur, X. Qiu, T. Parcollet, and N. D. Lane, “Flower: A friendly federated learning research framework,” CoRR, vol. abs/2007.14390, 2020. [Online]. Available: https://arxiv.org/abs/2007.14390

    2007 arXiv

  35. [44]

    Pyjoules: Python library that measures python code snippets,

    M. c. Belgaid, R. Rouvoy, and L. Seinturier, “Pyjoules: Python library that measures python code snippets,” Nov. 2019. [Online]. Available: https://github.com/powerapi-ng/pyJoules

    2019

  36. [2018]

    Lecture Notes in Computer Science, G

    Proceedings, ser. Lecture Notes in Computer Science, G. D. M. Serugendo and M. Loreti, Eds., vol. 10852. Springer, 2018, pp. 1–20. [Online]. Available: https://doi.org/10.1007/978-3-319-92408-3 1

    2018 doi

  37. [2025]

    Available: https://arxiv.org/abs/2503.20618

    [Online]. Available: https://arxiv.org/abs/2503.20618

    arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.