pith. sign in

arxiv: 2605.26323 · v3 · pith:J5FRDZFNnew · submitted 2026-05-25 · 💻 cs.DC

Totoro^+: An Adaptive and Scalable Edge Federated Learning System

Cheng-Wei Ching , Xin Chen , Taehwan Kim , Jian-Jhih Kuo , Dilma Da Silva , Liting Hu This is my paper

Pith reviewed 2026-06-29 20:06 UTC · model grok-4.3

classification 💻 cs.DC
keywords federated learningedge computingpeer-to-peerdecentralized systemsdistributed hash tablescalabilitynetwork churn
0
0 comments X

The pith

Totoro+ decentralizes federated learning by giving each application its own dedicated parameter server on edge nodes.

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

The paper establishes that a DHT-based peer-to-peer architecture can support many simultaneous federated learning applications on large edge networks without relying on a shared central server. By allowing any node to act in multiple roles such as coordinator or aggregator for any application, the system improves scalability and adaptivity to network changes and churn. Three innovations enable the design: a locality-aware P2P multi-ring structure, a publish/subscribe-based forest abstraction, and a game-theoretic path planning model that guarantees an epsilon-approximate Nash equilibrium. Experiments on 500 Amazon EC2 servers show the approach scales with the number of applications and nodes while delivering speedups and logarithmic communication costs.

Core claim

Totoro+ re-architects centralized FL into a fully decentralized P2P system using DHT where each FL application has its own dedicated parameter server, and any edge node can serve as coordinator, aggregator, client selector, or worker for any combination of applications, enabled by a locality-aware P2P multi-ring structure, a publish/subscribe-based forest abstraction, and a game-theoretic path planning model with epsilon-approximate Nash equilibrium guarantee.

What carries the argument

DHT-based peer-to-peer model with locality-aware multi-ring structure, publish/subscribe forest abstraction, and game-theoretic path planning for epsilon-approximate Nash equilibrium.

If this is right

  • The system scales gracefully with the number of FL applications and N edge nodes.
  • It speeds up the total training time by 1.2×-14.0×.
  • It achieves O(log N) hops for model dissemination and gradient aggregation with millions of nodes.
  • It efficiently adapts to practical edge networks and churns.

Where Pith is reading between the lines

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

  • The multi-role node design could reduce reliance on specialized infrastructure in dynamic edge settings.
  • The approach might extend to optimize additional metrics such as energy use during path planning.
  • Logarithmic scaling could support federated learning deployments in large-scale IoT networks with frequent node turnover.

Load-bearing premise

The game-theoretic path planning model delivers a guaranteed epsilon-approximate Nash equilibrium under realistic edge churn without new failure modes from multi-role nodes.

What would settle it

An experiment with high node churn rates where the path planning fails to maintain the approximate Nash equilibrium or causes training delays and failures.

Figures

Figures reproduced from arXiv: 2605.26323 by Cheng-Wei Ching, Dilma Da Silva, Jian-Jhih Kuo, Liting Hu, Taehwan Kim, Xin Chen.

Figure 1
Figure 1. Figure 1: An edge federated learning use case based on au [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The data pipeline in a typical FL framework. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The Totoro+ system overview. A. Overview Totoro+’s system goals are: • Scalability. Totoro+ can scale to process a vast number of FL applications’ tasks simultaneously on millions of edge nodes without introducing any centralized bottleneck. • Adaptivity. Totoro+ can quickly adapt to the practical edge networks characterized by varying bandwidths, unreliable links, high churn (nodes join and leave), and wo… view at source ↗
Figure 4
Figure 4. Figure 4: The workflow of Totoro+’s publish/subscribe-based “forest” abstraction with three dataflow trees for three FL applications and an advertise-discover tree for application advertise and discovery [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Totoro+ excels at scalability by fairly distributing many dynamic-structured dataflow trees across large-scale edge topologies. 20 40 80 160 320 640 1280 2560 5120 Number of Edge Nodes 0 1000 2000 3000 Model Dissemination Time (ms) Totoro (ResNet-34) Totoro+ (ResNet-34) Totoro (ShuffleNet V2) Totoro+ (ShuffleNet V2) (a) Model dissemination time. 20 40 80 160 320 640 1280 2560 5120 Number of Edge Nodes 0 10… view at source ↗
Figure 6
Figure 6. Figure 6: Totoro+ scales with an exponentially increasing number of edge nodes for model dissemination and gradient aggregation. 100 300 500 700 1000 Number of Dataflow Trees 1 1.19 2 4 8 10 Traffic Ratio per Node Totoro+ TCP Baselines TCP (a) TCP traffic cost. 100 300 500 700 1000 Number of Dataflow Trees 1 1.29 2 4 8 10 Traffic Ratio per Node Totoro+ UDP Baselines UDP (b) UDP traffic cost [PITH_FULL_IMAGE:figures… view at source ↗
Figure 7
Figure 7. Figure 7: Totoro+ reduces communication cost. each node in a 1000-node edge zone under stress testing. The results show that when we create a large number of dataflow trees like 500, 99.5% of the nodes are the roots of 3 trees or less. The results illustrate a good load balance among participating devices when performing a large number of FL applications’ tasks simultaneously [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Time-to-accuracy comparison of Totoro+, OpenFL, and FedScale. When 5∼20 applications’ models are simultaneously trained, Totoro+ speeds up the total training time 3.0×-14.0× to reach the equivalent model accuracy on the middle-scale Google Speech dataset. 0 3 6 9 12 Total Training Time (hours) 0 20 40 60 80 Accuracy (%) Totoro+ (f=8) Totoro+ (f=16) Totoro+ (f=32) OpenFL FedScale (a) 1 application’s model 0… view at source ↗
Figure 9
Figure 9. Figure 9: Time-to-accuracy comparison of Totoro+, OpenFL, and FedScale. When 5∼20 applications’ models are simultaneously trained, Totoro+ speeds up the total training time 1.2×-11.5× to reach the equivalent model accuracy on the large-scale FEMNIST dataset. connection. Then the overhead for creating new dataflow trees can be amortized over the overhead of the overlay. D. Federated learning effectiveness In this sec… view at source ↗
Figure 10
Figure 10. Figure 10: 100 edge servers and 900 end devices in Melbourne, [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗
Figure 14
Figure 14. Figure 14: Node selection frequencies generated by different algorithms. 0 200 400 600 800 1000 #Nodes 0 500 1000 1500 2000 Running Time (ms) Totoro Totoro+ [PITH_FULL_IMAGE:figures/full_fig_p014_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Running time comparison. 250 500 750 1000 #Nodes 0 25 50 75 100 Percentage (%) Line 5 Line 6 Line 7 Line 8 [PITH_FULL_IMAGE:figures/full_fig_p014_15.png] view at source ↗
Figure 20
Figure 20. Figure 20: Example of a dataflow tree. Algorithm 1. First, node l1 follows its initial policy π 1 l1 to select the next hops and collect rewards r 1,t n . Since there are two candidates for node l1 to forward its packets, initial policy π 1 l1 should be a 2-dimensional vector with the summation of its entries being 1. For example, if π 1 l1 = [0.5, 0.5], the chance of selecting either intermediate node is 50%. Suppo… view at source ↗
Figure 24
Figure 24. Figure 24: Cumulative packet sent with different τ . of packet transmission. Since latency is a continuous-valued metric and may exhibit large variance across different network conditions, we normalize it into a bounded reward signal to stabilize the learning process. In addition, the end-to-end latency experienced by a packet is inherently affected by the joint routing decisions of multiple nodes: when multiple nod… view at source ↗
Figure 26
Figure 26. Figure 26: Totoro+ keeps a high CPU uti￾lization in the node with more resource units. 3U 6U 10U 20U 40U Resource Units 0 1000 2000 3000 4000 Communication Time (s) Totoro+ (inter) Totoro+ (intra) Baseline (inter) Baseline (intra) 3U 6U 10U 20U 40U Resource Units 0 5 10 15 20 25 Number of Edges ( £1000) [PITH_FULL_IMAGE:figures/full_fig_p029_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: Totoro+ enables more intra￾and inter-node communication edges. 3U 6U 10U 20U 40U Resource Units 0 1000 2000 3000 4000 Communication Time (s) [PITH_FULL_IMAGE:figures/full_fig_p029_27.png] view at source ↗
Figure 29
Figure 29. Figure 29: Time-to-Accuracy performance. Decentralized federated learning frameworks running atop Totoro [PITH_FULL_IMAGE:figures/full_fig_p030_29.png] view at source ↗
read the original abstract

Federated Learning (FL) is an emerging distributed machine learning (ML) technique that enables in-situ model training and inference on decentralized edge devices. We propose Totoro$^+$, a novel scalable FL system that enables massive FL applications to run simultaneously on edge networks. The key insight is to explore a distributed hash table (DHT)-based peer-to-peer (P2P) model to re-architect the centralized FL system design into a fully decentralized one. In contrast to previous studies where many FL applications shared one centralized parameter server, Totoro$^+$ assigns a dedicated parameter server to each application. Any edge node can act as any application's coordinator, aggregator, client selector, worker (participant device), or any combination of the above, thereby radically improving scalability and adaptivity. Totoro$^+$ introduces three innovations to realize its design: a locality-aware P2P multi-ring structure, a publish/subscribe-based forest abstraction, and a game-theoretic path planning model with a guarantee of an $\epsilon$-approximate Nash equilibrium. Real-world experiments on 500 Amazon EC2 servers show that Totoro$^+$ scales gracefully with the number of FL applications and $N$ edge nodes speeds up the total training time by $1.2\times-14.0\times$, achieves $\mathcal{O}(\log N)$ hops for model dissemination and gradient aggregation with millions of nodes, and efficiently adapts to the practical edge networks and churns.

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.

Referee Report

3 major / 1 minor

Summary. The manuscript proposes Totoro+, a fully decentralized edge federated learning system that replaces centralized parameter servers with a DHT-based P2P architecture. Each FL application receives a dedicated parameter server, and any edge node can dynamically assume roles including coordinator, aggregator, client selector, or worker. The design rests on three innovations: a locality-aware P2P multi-ring overlay, a publish/subscribe forest abstraction, and a game-theoretic path planning model asserted to deliver an ε-approximate Nash equilibrium. Experiments on 500 Amazon EC2 servers report 1.2×–14.0× speedups in total training time, O(log N) hops for dissemination and aggregation (extrapolated to millions of nodes), and adaptation to churn.

Significance. If the O(log N) bound, speedups, and ε-Nash guarantee are rigorously established, the work would meaningfully advance scalable concurrent FL on edge networks by removing single-point bottlenecks and providing a formal handle on dynamic role assignment.

major comments (3)
  1. [Abstract] Abstract: the O(log N) hop bound and million-node scalability claim rest on 500-node EC2 runs; no analytic derivation, larger-scale simulation, or measurement of per-node state growth is supplied to justify the extrapolation.
  2. [Abstract] Abstract (game-theoretic path planning model): the ε-approximate Nash equilibrium is asserted to hold while nodes dynamically act as coordinators/aggregators under churn, yet the 500-node experiments cannot exercise equilibrium computation cost or re-convergence after churn events at the claimed million-node regime; if equilibrium maintenance incurs super-logarithmic overhead, both the hop bound and reported speedups become unsupported.
  3. [Abstract] Abstract: the claim that any node can reliably serve multiple roles without introducing new failure modes or security issues is central to the adaptivity argument, but no analysis of coordinator/aggregator load imbalance or churn-induced re-election latency is provided.
minor comments (1)
  1. The abstract refers to 'practical edge networks and churns' without defining quantitative metrics (e.g., churn rate, latency distribution) or baselines used to evaluate adaptation efficiency.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below, clarifying the foundations of our claims while committing to revisions that strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the O(log N) hop bound and million-node scalability claim rest on 500-node EC2 runs; no analytic derivation, larger-scale simulation, or measurement of per-node state growth is supplied to justify the extrapolation.

    Authors: The O(log N) bound is inherited from the standard DHT properties of the multi-ring overlay (analogous to Chord), which have been analytically proven in prior literature; our locality-aware extension preserves the bound while adding edge-specific optimizations. The 500-node experiments confirm the bound in practice and show graceful scaling. To directly address the extrapolation, we will add (i) a concise analytic derivation of hop count and state growth, (ii) per-node state measurements, and (iii) results from larger-scale simulations (up to 10k nodes) in the revised manuscript. revision: yes

  2. Referee: [Abstract] Abstract (game-theoretic path planning model): the ε-approximate Nash equilibrium is asserted to hold while nodes dynamically act as coordinators/aggregators under churn, yet the 500-node experiments cannot exercise equilibrium computation cost or re-convergence after churn events at the claimed million-node regime; if equilibrium maintenance incurs super-logarithmic overhead, both the hop bound and reported speedups become unsupported.

    Authors: The game-theoretic path planner computes an ε-approximate Nash equilibrium locally within small node groups using a distributed algorithm whose per-decision cost is bounded by O(log N). Experiments already demonstrate stable churn adaptation on 500 nodes. We agree that explicit complexity analysis and larger-scale re-convergence data would strengthen the claims; we will add both a formal complexity argument and additional simulation results showing equilibrium maintenance overhead remains logarithmic under churn. revision: yes

  3. Referee: [Abstract] Abstract: the claim that any node can reliably serve multiple roles without introducing new failure modes or security issues is central to the adaptivity argument, but no analysis of coordinator/aggregator load imbalance or churn-induced re-election latency is provided.

    Authors: Role flexibility is enabled by the DHT and pub/sub forest, which distribute responsibilities and inherit load-balancing properties from the overlay. Our experiments already measure adaptation under churn. We will add an explicit analysis section that reports observed load-imbalance metrics, re-election latencies, and a discussion of potential failure modes. A brief treatment of inherited DHT security properties and mitigation strategies will also be included. revision: yes

Circularity Check

0 steps flagged

No circularity: claims rest on independent architectural design and external benchmarks

full rationale

The paper presents three explicit innovations (locality-aware P2P multi-ring, pub/sub forest, game-theoretic path planning) whose performance claims are evaluated via 500-node EC2 experiments and standard DHT properties for O(log N) hops. No equations, fitted parameters, or self-referential definitions appear in the abstract or described structure; the ε-Nash guarantee is asserted as a model property rather than derived from the reported speedups or hop counts. The derivation chain is therefore self-contained against external benchmarks and does not reduce any prediction to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, invented physical entities, or non-standard axioms; the three innovations are presented as engineering constructs rather than new theoretical primitives.

axioms (1)
  • domain assumption Edge nodes can reliably perform multiple FL roles (coordinator, aggregator, worker) under churn
    Invoked to support the claim that any node can act as any role for any application.

pith-pipeline@v0.9.1-grok · 5805 in / 1325 out tokens · 29518 ms · 2026-06-29T20:06:22.465936+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

139 extracted references · 4 canonical work pages

  1. [1]

    Totoro: A scalable federated learning engine for the edge,

    C.-W. Ching, X. Chen, T. Kim, B. Ji, Q. Wang, D. Da Silva, and L. Hu, “Totoro: A scalable federated learning engine for the edge,” inProceedings of the Nineteenth European Conference on Computer Systems, 2024, p. 182–199. [Online]. Available: https://doi.org/10.1145/3627703.3629575

    work page doi:10.1145/3627703.3629575 2024

  2. [2]

    Asynchronous online federated learning for edge devices with non-iid data,

    Y . Chen, Y . Ning, M. Slawski, and H. Rangwala, “Asynchronous online federated learning for edge devices with non-iid data,” inIEEE International Conference on Big Data, 2020, pp. 15–24

    2020

  3. [3]

    Communication-efficient federated deep learning with layerwise asynchronous model update and temporally weighted aggregation,

    Y . Chen, X. Sun, and Y . Jin, “Communication-efficient federated deep learning with layerwise asynchronous model update and temporally weighted aggregation,”IEEE Transactions on Neural Networks and Learning Systems, vol. 31, no. 10, pp. 4229–4238, 2020

    2020

  4. [4]

    Federated multi-task learning,

    V . Smith, C.-K. Chiang, M. Sanjabi, and A. S. Talwalkar, “Federated multi-task learning,”Advances in neural information processing sys- tems, vol. 30, 2017

    2017

  5. [5]

    Federated learning for mobile keyboard prediction,

    A. Hardet al., “Federated learning for mobile keyboard prediction,” arXiv preprint arXiv:1811.03604, 2018

    Pith/arXiv arXiv 2018

  6. [6]

    Applied federated learning: Improving google keyboard query suggestions,

    T. Yanget al., “Applied federated learning: Improving google keyboard query suggestions,”arXiv preprint arXiv:1812.02903, 2018

    Pith/arXiv arXiv 2018

  7. [7]

    IBM federated learning: an enterprise framework white paper v0. 1,

    H. Ludwiget al., “IBM federated learning: an enterprise framework white paper v0. 1,”arXiv preprint arXiv:2007.10987, 2020

    arXiv 2007

  8. [8]

    https: //www.tensorflow.org/federated

    Tensorflow federated: Machine learning on decentralized data. https: //www.tensorflow.org/federated

  9. [9]

    Leaf: A benchmark for federated settings,

    S. Caldaset al., “Leaf: A benchmark for federated settings,”arXiv preprint arXiv:1812.01097, 2018

    arXiv 2018

  10. [10]

    FLOAT: Federated learning optimizations with auto- mated tuning,

    A. F. Khan, A. A. Khan, A. M. Abdelmoniem, S. Fountain, A. R. Butt, and A. Anwar, “FLOAT: Federated learning optimizations with auto- mated tuning,” inProceedings of the Nineteenth European Conference on Computer Systems, 2024, pp. 200–218

    2024

  11. [11]

    REFL: Resource-efficient federated learning,

    A. M. Abdelmoniem, A. N. Sahu, M. Canini, and S. A. Fahmy, “REFL: Resource-efficient federated learning,” inProceedings of the Eighteenth European Conference on Computer Systems, 2023, pp. 215–232

    2023

  12. [12]

    https://github.com/ OpenMined/PySyft

    PySyft: A library for easy federated learning. https://github.com/ OpenMined/PySyft

  13. [13]

    United states department of transportation annual modal research plans,

    “United states department of transportation annual modal research plans,” https://www.transportation.gov/sites/dot.gov/files/2024-08/ AMRP%20FY2024%20-%202025%20ITSJPO.pdf

    2024

  14. [14]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,”Neural Computation, vol. 9, no. 8, pp. 1735–1780, 1997

    1997

  15. [15]

    Driver monitoring-based lane-change prediction: A personalized federated learning framework,

    R. Du, K. Han, R. Gupta, S. Chen, S. Labi, and Z. Wang, “Driver monitoring-based lane-change prediction: A personalized federated learning framework,” inIEEE Intelligent Vehicles Symposium, 2023, pp. 1–7

    2023

  16. [16]

    Fedstream: A federated learn- ing framework on heterogeneous streaming data for next-generation traffic analysis,

    N. Wang, X. Li, Z. Guan, and S. Yuan, “Fedstream: A federated learn- ing framework on heterogeneous streaming data for next-generation traffic analysis,”IEEE Transactions on Network Science and Engineer- ing, vol. 11, no. 3, pp. 2485–2496, 2023

    2023

  17. [17]

    Ai- empowered trajectory anomaly detection for intelligent transportation systems: A hierarchical federated learning approach,

    X. Wang, W. Liu, H. Lin, J. Hu, K. Kaur, and M. S. Hossain, “Ai- empowered trajectory anomaly detection for intelligent transportation systems: A hierarchical federated learning approach,”IEEE Transac- tions on Intelligent Transportation Systems, vol. 24, no. 4, pp. 4631– 4640, 2022

    2022

  18. [18]

    http://www.bittorrent.com/

    Bittorrent. http://www.bittorrent.com/

  19. [19]

    http://storj.io/

    Storj.io. http://storj.io/

  20. [20]

    https://freenetproject.org/

    Freenet: The free network. https://freenetproject.org/

  21. [21]

    Peer-assisted content distribution in akamai netses- sion,

    M. Zhaoet al., “Peer-assisted content distribution in akamai netses- sion,” inProceedings of the 2013 conference on Internet measurement conference, 2013, pp. 31–42

    2013

  22. [22]

    On scaling decentralized blockchains: (a position paper),

    K. Cromanet al., “On scaling decentralized blockchains: (a position paper),” inInternational conference on financial cryptography and data security. Springer, 2016, pp. 106–125

    2016

  23. [23]

    https://www.freepastry.org/FreePastry/

    Pastry. https://www.freepastry.org/FreePastry/

  24. [24]

    PyTorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, “PyTorch: An imperative style, high-performance deep learning library,” inAdvances in Neural Information Processing Sy...

  25. [25]

    Available: https://proceedings.neurips.cc/paper files/ paper/2019/file/bdbca288fee7f92f2bfa9f7012727740-Paper.pdf

    [Online]. Available: https://proceedings.neurips.cc/paper files/ paper/2019/file/bdbca288fee7f92f2bfa9f7012727740-Paper.pdf

    2019

  26. [26]

    https:// openfl.readthedocs.io/

    OpenFL - an open-source framework for federated learning. https:// openfl.readthedocs.io/

  27. [27]

    FedScale: Benchmarking model and system performance of federated learning at scale,

    F. Lai, Y . Dai, S. Singapuram, J. Liu, X. Zhu, H. Madhyastha, and M. Chowdhury, “FedScale: Benchmarking model and system performance of federated learning at scale,” inProceedings of the 39th International Conference on Machine Learning, vol. 162. PMLR, 17–23 Jul 2022, pp. 11 814–11 827. [Online]. Available: https://proceedings.mlr.press/v162/lai22a.html

    2022

  28. [28]

    A crowdsourcing framework for on-device federated learning,

    S. R. Pandey, N. H. Tran, M. Bennis, Y . K. Tun, A. Manzoor, and C. S. Hong, “A crowdsourcing framework for on-device federated learning,” IEEE Transactions on Wireless Communications, vol. 19, no. 5, pp. 3241–3256, 2020

    2020

  29. [29]

    CoPiFL: A collusion-resistant and privacy-preserving federated learning crowdsourcing scheme using blockchain and homo- morphic encryption,

    R. Xionget al., “CoPiFL: A collusion-resistant and privacy-preserving federated learning crowdsourcing scheme using blockchain and homo- morphic encryption,”Future Generation Computer Systems, vol. 156, pp. 95–104, 2024

    2024

  30. [30]

    PAPAY A: Practical, private, and scalable federated learning,

    D. Hubaet al., “PAPAY A: Practical, private, and scalable federated learning,”Proceedings of Machine Learning and Systems, vol. 4, pp. 814–832, 2022

    2022

  31. [31]

    FLINT: A platform for federated learning integration,

    E. Wang, B. Chen, M. Chowdhury, A. Kannan, and F. Liang, “FLINT: A platform for federated learning integration,”Proceedings of Machine Learning and Systems, vol. 5, pp. 21–34, 2023

    2023

  32. [32]

    Towards federated learning at scale: System design,

    K. Bonawitzet al., “Towards federated learning at scale: System design,” inProceedings of Machine Learning and Systems, vol. 1, 2019, pp. 374–388

    2019

  33. [33]

    Federated evaluation and tuning for on-device personalization: System design & applications,

    M. Pauliket al., “Federated evaluation and tuning for on-device personalization: System design & applications,”arXiv preprint arXiv:2102.08503, 2021

    arXiv 2021

  34. [34]

    Testing the resilience of mec-based iot applications against resource exhaustion attacks,

    R. Pietrantuono, M. Ficco, and F. Palmieri, “Testing the resilience of mec-based iot applications against resource exhaustion attacks,”IEEE Transactions on Dependable and Secure Computing, vol. 21, no. 2, pp. 804–818, 2024

    2024

  35. [35]

    TiFL: A tier-based federated learning system,

    Z. Chaiet al., “TiFL: A tier-based federated learning system,” in Proceedings of the 29th international symposium on high-performance parallel and distributed computing, 2020, pp. 125–136

    2020

  36. [36]

    Oort: Efficient federated learning via guided participant selection,

    F. Lai, X. Zhu, H. V . Madhyastha, and M. Chowdhury, “Oort: Efficient federated learning via guided participant selection,” in15th USENIX Symposium on Operating Systems Design and Implementation, 2021, pp. 19–35

    2021

  37. [37]

    Client selection for federated learning with heterogeneous resources in mobile edge,

    T. Nishio and R. Yonetani, “Client selection for federated learning with heterogeneous resources in mobile edge,” inIEEE International Conference on Communications, 2019, pp. 1–7

    2019

  38. [38]

    QSGD: Communication-efficient sgd via gradient quantization and encoding,

    D. Alistarh, D. Grubic, J. Li, R. Tomioka, and M. V ojnovic, “QSGD: Communication-efficient sgd via gradient quantization and encoding,” Advances in neural information processing systems, vol. 30, 2017. 17

    2017

  39. [39]

    signSGD with majority vote is communication efficient and fault tolerant,

    J. Bernstein, J. Zhao, K. Azizzadenesheli, and A. Anandkumar, “signSGD with majority vote is communication efficient and fault tolerant,” inInternational Conference on Learning Representations, 2019

    2019

  40. [40]

    Communication-efficient learning of deep networks from decentral- ized data,

    B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentral- ized data,” inArtificial intelligence and statistics. PMLR, 2017, pp. 1273–1282

    2017

  41. [41]

    Semi-synchronous federated learning for energy-efficient training and accelerated conver- gence in cross-silo settings,

    D. Stripelis, P. M. Thompson, and J. L. Ambite, “Semi-synchronous federated learning for energy-efficient training and accelerated conver- gence in cross-silo settings,”ACM Transactions on Intelligent Systems and Technology, vol. 13, no. 5, pp. 1–29, 2022

    2022

  42. [42]

    Asynchronous federated optimiza- tion,

    C. Xie, S. Koyejo, and I. Gupta, “Asynchronous federated optimiza- tion,” inAdvances in Neural Information Processing Systems Workshop on Optimization for Machine Learning, 2020

    2020

  43. [43]

    Adaptive federated learning in resource constrained edge computing systems,

    S. Wanget al., “Adaptive federated learning in resource constrained edge computing systems,”IEEE Journal on Selected Areas in Commu- nications, vol. 37, no. 6, pp. 1205–1221, 2019

    2019

  44. [44]

    Federated learning for edge networks: Resource optimization and incentive mechanism,

    L. U. Khanet al., “Federated learning for edge networks: Resource optimization and incentive mechanism,”IEEE Communications Mag- azine, vol. 58, no. 10, pp. 88–93, 2020

    2020

  45. [45]

    Serverless edge computing: vision and challenges,

    M. S. Aslanpouret al., “Serverless edge computing: vision and challenges,” inProceedings of the 2021 Australasian computer science week multiconference, 2021, pp. 1–10

    2021

  46. [46]

    Edge-assisted democratized learning toward federated analytics,

    S. R. Pandeyet al., “Edge-assisted democratized learning toward federated analytics,”IEEE Internet of Things Journal, vol. 9, no. 1, pp. 572–588, 2022

    2022

  47. [47]

    AgileDART: An agile and scalable edge stream processing engine,

    C.-W. Ching, X. Chen, C. Kim, T. Wang, D. Chen, D. Da Silva, and L. Hu, “AgileDART: An agile and scalable edge stream processing engine,”IEEE Transactions on Mobile Computing, vol. 24, no. 5, pp. 4510–4528, 2025

    2025

  48. [48]

    Kalmia: A heterogeneous qos-aware scheduling framework for dnn tasks on edge servers,

    Z. Fu, J. Ren, D. Zhang, Y . Zhou, and Y . Zhang, “Kalmia: A heterogeneous qos-aware scheduling framework for dnn tasks on edge servers,” inConference on Computer Communications, 2022, pp. 780– 789

    2022

  49. [49]

    Learning in congestion games with bandit feedback,

    Q. Cui, Z. Xiong, M. Fazel, and S. S. Du, “Learning in congestion games with bandit feedback,”Advances in Neural Information Pro- cessing Systems, vol. 35, pp. 11 009–11 022, 2022

    2022

  50. [50]

    Pastry: Scalable, decentralized object lo- cation, and routing for large-scale peer-to-peer systems,

    A. Rowstron and P. Druschel, “Pastry: Scalable, decentralized object lo- cation, and routing for large-scale peer-to-peer systems,” inIFIP/ACM International Conference on Distributed Systems Platforms. Springer, 2001, pp. 329–350

    2001

  51. [51]

    Chord: A scalable peer-to-peer lookup service for internet applica- tions,

    I. Stoica, R. Morris, D. Karger, M. F. Kaashoek, and H. Balakrishnan, “Chord: A scalable peer-to-peer lookup service for internet applica- tions,”ACM SIGCOMM computer communication review, vol. 31, no. 4, pp. 149–160, 2001

    2001

  52. [52]

    Tapestry: a resilient global-scale overlay for service deployment,

    B. Zhao, L. Huang, J. Stribling, S. Rhea, A. Joseph, and J. Kubiatowicz, “Tapestry: a resilient global-scale overlay for service deployment,” IEEE Journal on Selected Areas in Communications, vol. 22, no. 1, pp. 41–53, 2004

    2004

  53. [53]

    Fedhealth: A federated transfer learning framework for wearable healthcare,

    Y . Chen, X. Qin, J. Wang, C. Yu, and W. Gao, “Fedhealth: A federated transfer learning framework for wearable healthcare,”IEEE Intelligent Systems, vol. 35, no. 4, pp. 83–93, 2020

    2020

  54. [54]

    Fedsens: A federated learning approach for smart health sensing with class imbalance in resource constrained edge computing,

    D. Y . Zhang, Z. Kou, and D. Wang, “Fedsens: A federated learning approach for smart health sensing with class imbalance in resource constrained edge computing,” inIEEE Conference on Computer Com- munications, 2021, pp. 1–10

    2021

  55. [55]

    Kademlia: A peer-to-peer informa- tion system based on the xor metric,

    P. Maymounkov and D. Mazieres, “Kademlia: A peer-to-peer informa- tion system based on the xor metric,” inInternational workshop on peer-to-peer systems. Springer, 2002, pp. 53–65

    2002

  56. [56]

    Topologically-aware overlay construction and server selection,

    S. Ratnasamy, M. Handley, R. Karp, and S. Shenker, “Topologically-aware overlay construction and server selection,” inProceedings.Twenty-First Annual Joint Conference of the IEEE Computer and Communications Societies, vol. 3, 2002, pp. 1190–1199 vol.3

    2002

  57. [57]

    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

    2020

  58. [58]

    Differentially private federated learning: A client level perspective,

    R. C. Geyer, T. Klein, and M. Nabi, “Differentially private federated learning: A client level perspective,” inAdvances in Neural Information Processing Systems Workshop: Machine Learning on the Phone and other Consumer Devices, 2017

    2017

  59. [59]

    BatchCrypt: Efficient homomorphic encryption for Cross-Silo federated learning,

    C. Zhang, S. Li, J. Xia, W. Wang, F. Yan, and Y . Liu, “BatchCrypt: Efficient homomorphic encryption for Cross-Silo federated learning,” inUSENIX annual technical conference, 2020, pp. 493–506

    2020

  60. [60]

    Joint con- figuration adaptation and bandwidth allocation for edge-based real-time video analytics,

    C. Wang, S. Zhang, Y . Chen, Z. Qian, J. Wu, and M. Xiao, “Joint con- figuration adaptation and bandwidth allocation for edge-based real-time video analytics,” inIEEE Conference on Computer Communications, 2020, pp. 257–266

    2020

  61. [61]

    Stochastic online shortest path routing: The value of feedback,

    M. S. Talebi, Z. Zou, R. Combes, A. Proutiere, and M. Johansson, “Stochastic online shortest path routing: The value of feedback,”IEEE Transactions on Automatic Control, vol. 63, no. 4, pp. 915–930, 2018

    2018

  62. [62]

    Regret analysis of stochastic and nonstochastic multi-armed bandit problems,

    S. Bubecket al., “Regret analysis of stochastic and nonstochastic multi-armed bandit problems,”Foundations and Trends® in Machine Learning, vol. 5, no. 1, pp. 1–122, 2012

    2012

  63. [63]

    Algorithmic game theory,

    T. Roughgarden, “Algorithmic game theory,”Communications of the ACM, vol. 53, no. 7, pp. 78–86, 2010

    2010

  64. [64]

    Independent policy gradient for large-scale Markov potential games: Sharper rates, func- tion approximation, and game-agnostic convergence,

    D. Ding, C.-Y . Wei, K. Zhang, and M. Jovanovic, “Independent policy gradient for large-scale Markov potential games: Sharper rates, func- tion approximation, and game-agnostic convergence,” inInternational Conference on Machine Learning. PMLR, 2022, pp. 5166–5220

    2022

  65. [65]

    Learning to compete, compromise, and cooperate in repeated general-sum games,

    J. W. Crandall and M. A. Goodrich, “Learning to compete, compromise, and cooperate in repeated general-sum games,” in Proceedings of the 22nd International Conference on Machine Learning, 2005, p. 161–168. [Online]. Available: https://doi.org/10. 1145/1102351.1102372

    arXiv 2005

  66. [66]

    Policy gradient methods find the nash equilibrium in n-player general-sum linear-quadratic games,

    B. Hambly, R. Xu, and H. Yang, “Policy gradient methods find the nash equilibrium in n-player general-sum linear-quadratic games,” Journal of Machine Learning Research, vol. 24, no. 139, pp. 1–56,

  67. [67]

    Available: http://jmlr.org/papers/v24/21-0842.html

    [Online]. Available: http://jmlr.org/papers/v24/21-0842.html

  68. [68]

    Lattimore and C

    T. Lattimore and C. Szepesv ´ari,Bandit algorithms. Cambridge University Press, 2020

    2020

  69. [69]

    Online influence maxi- mization under independent cascade model with semi-bandit feedback,

    Z. Wen, B. Kveton, M. Valko, and S. Vaswani, “Online influence maxi- mization under independent cascade model with semi-bandit feedback,” Advances in neural information processing systems, vol. 30, 2017

    2017

  70. [70]

    An algorithm for quadratic programming,

    M. Franket al., “An algorithm for quadratic programming,”Naval research logistics quarterly, vol. 3, no. 1-2, pp. 95–110, 1956

    1956

  71. [71]

    https://pytorch.org/

    PyTorch. https://pytorch.org/

  72. [72]

    Scribe: a large-scale and decentralized application-level multicast infrastructure,

    M. Castro, P. Druschel, A.-M. Kermarrec, and A. Rowstron, “Scribe: a large-scale and decentralized application-level multicast infrastructure,” IEEE Journal on Selected Areas in Communications, vol. 20, no. 8, pp. 1489–1499, 2002

    2002

  73. [73]

    https://docs.pytorch.org/vision/stable/index.html

    torchvision. https://docs.pytorch.org/vision/stable/index.html

  74. [74]

    https://huggingface.co/docs/transformers/ en/index

    Hugging Face Transformers. https://huggingface.co/docs/transformers/ en/index

  75. [75]

    Optimal edge user allocation in edge computing with variable sized vector bin packing,

    P. Laiet al., “Optimal edge user allocation in edge computing with variable sized vector bin packing,” in16th International Conference on Service-Oriented Computing. Springer, 2018, pp. 230–245

    2018

  76. [76]

    Symbiotic lab,

    “Symbiotic lab,” https://symbioticlab.org/

  77. [77]

    Speech commands: A dataset for limited-vocabulary speech recognition,

    P. Warden, “Speech commands: A dataset for limited-vocabulary speech recognition,”arXiv preprint arXiv:1804.03209, 2018

    Pith/arXiv arXiv 2018

  78. [78]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, June 2016

    2016

  79. [79]

    ShuffleNet V2: Practical guidelines for efficient cnn architecture design,

    N. Ma, X. Zhang, H.-T. Zheng, and J. Sun, “ShuffleNet V2: Practical guidelines for efficient cnn architecture design,” inProceedings of the European Conference on Computer Vision, September 2018

    2018

  80. [80]

    AWStream: Adaptive wide-area streaming analytics,

    B. Zhang, X. Jin, S. Ratnasamy, J. Wawrzynek, and E. A. Lee, “AWStream: Adaptive wide-area streaming analytics,” inProceedings of the 2018 Conference of the ACM Special Interest Group on Data Communication, 2018, pp. 236–252

    2018

Showing first 80 references.