Pith. sign in

REVIEW 2 major objections 5 minor 69 references

Decentralised federated learning mixes like lazy random walks on temporal networks, so real-world heterogeneities slow convergence by tens to more than a hundred times relative to the usual synthetic setups.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-12 04:21 UTC pith:5MNQPRRF

load-bearing objection Clean continuous-time mapping of DFL averaging onto lazy RW diffusion, plus careful microcanonical evidence that real contact heterogeneities slow mixing by 1–2 orders of magnitude relative to the usual ER+regular benchmarks. the 2 major comments →

arxiv 2607.03171 v1 pith:5MNQPRRF submitted 2026-07-03 cs.LG cs.AIcs.DC

Decentralised Federated Learning over Temporal Networks: The Role of Heterogeneities

Arash Badie-Modiri , Chiara Boldrini , Lorenzo Valerio , J\'anos Kert\'esz , M\'arton Karsai This is my paper
classification cs.LG cs.AIcs.DC
keywords decentralised federated learningtemporal networkslazy random walkdiffusionheterogeneityinverse participation ratiopeer-to-peer aggregationmixing time
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper argues that when devices train a shared model by averaging parameters with whoever they meet, the speed of agreement is not set by the learning rule so much as by how information diffuses across the contact network. That diffusion is the same as a lazy random walk on a temporal network, both while models are still aligning and later when local training steps act as small kicks that then spread. Standard experiments use regular graphs and regular schedules and therefore paint a picture of unrealistically fast mixing. Real contact traces—school face-to-face, urban taxis, campus Wi-Fi—contain spatial embedding, bursty and self-exciting timing, and heterogeneous activity windows; each of those features slows diffusion, and together they can make a fully randomised network with the same size and event count mix tens to more than a hundred times faster than the original. The practical upshot is that protocol benchmarks that ignore these heterogeneities systematically overestimate how quickly decentralised learning will converge on real devices.

Core claim

The dissemination of model parameters under local averaging is governed, both in the early synchronisation phase and in the late stationary regime, by the same dynamics as a lazy random-walk diffusion process on the underlying temporal network. Structural and temporal heterogeneities universally slow that process, so that the homogeneous random-graph, regular-interval experiments common in the literature overestimate convergence speed by one to two orders of magnitude relative to empirical contact networks.

What carries the argument

The isomorphism between pairwise averaging and a time-reversed lazy random walk: the influence of each node’s initial parameters (or of a later local-training impulse) equals the visit probability of a walker that, at every contact, moves or stays with equal probability. Mixing speed is then read from the decay of the inverse participation ratio of those probabilities, which is proportional to parameter variance across nodes.

Load-bearing premise

Aggregation is instantaneous pairwise averaging (or a rule that becomes linear when models are already close) on single contacts with unlimited bandwidth and no simultaneous multi-neighbour meetings; local training steps are then treated as independent impulses that simply superpose under the same diffusion operator.

What would settle it

On any of the three empirical traces, measure the inverse-participation-ratio decay (or the cross-node parameter variance) under true pairwise averaging and under a fully randomised network that keeps the same nodes, links and total events; if the randomised network is not tens to more than a hundred times faster, or if the measured trajectories deviate systematically from the predicted lazy-walk kernel once local training is included as additive impulses, the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The manuscript maps decentralised federated learning (DFL) with local pairwise averaging over a continuous-time dyadic contact network to lazy random-walk diffusion on temporal networks. In the early phase, influence coefficients of initial parameters equal reverse-time lazy-walk visit probabilities (Eqs. 3–6), so node-parameter variance tracks the inverse participation ratio (IPR). In the stationary regime, injected learning perturbations propagate under the continuous-time Laplacian of the contact process (Eqs. 7–8), with a first-order linearisation for DecDiff-type rules. Using this proxy, the authors show that spatial embedding, heavy-tailed inter-event times, and self-excitation each slow IPR decay on synthetic models, and that microcanonical randomisations of three real contact traces (high-school RFID, SF cabs, KTH Wi-Fi) accelerate mixing by large factors relative to the empirical networks—implying that the common DFL benchmark (homogeneous random graphs with regular intervals) systematically overestimates convergence speed.

Significance. If the mapping and the measured slowdowns hold, the paper supplies a concrete, network-science-grounded reason why standard DFL simulation setups are optimistically biased, and a diagnostic (IPR decay under lazy temporal walks) that can be used without full end-to-end training. Strengths include a first-principles combinatorial derivation of the early-phase equivalence, an explicit Laplacian characterisation of stationary response, isolation of heterogeneity classes via microcanonical null models (Table II), and additive-impulse validation of the diffusion kernel on MNIST (Appendix A, Figs. 5–6). The work is a useful bridge between DFL protocol evaluation and temporal-network diffusion theory, with direct implications for benchmark design.

major comments (2)
  1. The headline quantitative claim that a fully randomised baseline “can mix tens to more than a hundred times faster” (Introduction findings; Conclusion) is not backed by an explicit table of mixing-time (or IPR half-life / 1/e) ratios for the three empirical networks versus Link, Timeline, and Link+Timeline. Fig. 4 shows order-of-magnitude IPR gaps, but the numerical factors should be reported so the claim is falsifiable and comparable across datasets.
  2. Section IV-B and Appendix A treat local-learning updates as independent impulses that superpose under the same Laplacian. The manuscript correctly notes that generation of increments is state-dependent and non-i.i.d., yet the only end-to-end check is on a small ER graph with shared MNIST data (Figs. 5–6). A short discussion or experiment under non-IID local data (or a clear statement that the diffusion claim is about propagation only) would better bound when the stationary approximation remains predictive for realistic DFL.
minor comments (5)
  1. Fig. 3 panels (c)–(d) and Fig. 4 use “p2_i - 1/n” / “p2_i 1/n” in axis labels; standardise to IPR notation (e.g. ∑_i P_i²(t) − 1/N) and ensure the −1/N offset is applied consistently in all panels.
  2. Section III: the assumption that simultaneous multi-neighbour events are resolved sequentially in random order is stated, but a one-sentence pointer to the hypergraph generalisation (Eq. 6) would help readers who expect multi-way aggregation.
  3. Table I reports time windows and event counts after pre-processing; briefly note how sensitive the IPR trajectories are to the 3/4-window strong-connectivity filter, or that results are robust under modest changes of that threshold.
  4. Related work: a few recent DFL-over-time-varying-graph papers are cited; a short explicit contrast with switching-topology consensus (joint connectivity windows) would clarify what is new relative to classical product-of-stochastic-matrices results.
  5. Typos / polish: “inhomogeneities” vs “heterogeneities” is used interchangeably in abstract and body—pick one; “DecDiff” update (Eq. 9) has a typesetting glitch around the norm term that should be cleaned for production.

Circularity Check

0 steps flagged

No significant circularity: the lazy-RW mapping follows directly from pairwise averaging by construction, and heterogeneity slowdowns are measured against independent microcanonical nulls.

full rationale

The early-phase influence coefficients a_ij(t) are defined combinatorially as the sum over reverse time-respecting paths of 2^{-|inc(p)|} (Eqs. 3-4), which is identically the visit probability of a reverse lazy random walk under the paper's own aggregation rule (Eq. 2). The stationary response is likewise the continuous-time Laplacian evolution E[dw] = -½ L_λ w dt (Eqs. 7-8) obtained by writing the expected drift of pairwise averaging. Both equalities are therefore tautological under the stated model assumptions rather than circular predictions. The DecDiff linearisation (around Eq. 12) and the additive-impulse checks in Appendix A are validations, not fits. Prior static-graph results [11] by overlapping authors are cited only as motivation and are re-derived and extended; they are not load-bearing uniqueness theorems. Empirical claims rest on synthetic ensembles and microcanonical randomisations that destroy the heterogeneities under study, furnishing independent contrasts. No parameter is fitted and then re-presented as a prediction, and no known empirical pattern is merely renamed. Score 1 reflects only the non-load-bearing self-citation.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 0 invented entities

The central claim rests on standard random-walk and Laplacian theory plus a small set of modelling choices that define the DFL setting under study. No free parameters are fitted to produce the slowdown claim; the synthetic parameters (γ, α, T, d) are varied only to isolate mechanisms. No new physical entities are postulated.

axioms (4)
  • domain assumption Pairwise averaging (or any aggregation that reduces to a weighted linear average in the small-discrepancy regime) is the aggregation rule; simultaneous multi-neighbour events are resolved sequentially at random.
    Stated in Section III, Eqs. (1)–(2) and the paragraph following Eq. (6); required for the influence coefficients to equal reverse lazy-walk probabilities.
  • domain assumption Contacts are instantaneous (zero duration) and bandwidth is unbounded; medium contention is neglected.
    Explicit modelling choice in Section III for mathematical tractability; prolonged contacts are only approximated by periodic pings in the experiments.
  • domain assumption In the early phase aggregation dominates local learning; in the stationary phase learning increments act as independent impulses that superpose linearly under the same diffusion operator.
    Section IV-A and IV-B; the linear-superposition claim is validated numerically but not proved for arbitrary non-linear SGD.
  • standard math Standard continuous-time lazy random-walk theory on temporal networks (mixing, IPR, Laplacian generators) applies without modification.
    Used throughout Section IV; citations to the temporal-network literature supply the background results.

pith-pipeline@v1.1.0-grok45 · 29008 in / 2608 out tokens · 28019 ms · 2026-07-12T04:21:32.306671+00:00 · methodology

0 comments
read the original abstract

Decentralised federated learning, based on peer-to-peer communication, is increasingly proposed for on-device training of machine learning models, promising a privacy-preserving, communication-efficient training process with no risk of single-point failure. However, the role of structural and temporal inhomogeneities in such fully decentralised settings remains poorly understood. Here, we investigate their effects when model parameters are locally averaged during aggregation. We show that the decentralised federated learning process is governed, both in the early phase and the late, stationary limit, by the same dynamics as a lazy random-walk diffusion process on temporal networks. Based on this mapping, we demonstrate that the typical experimental scenario used in decentralised federated learning leads to unrealistically rapid convergence because of ignoring the temporal and structural inhomogeneities inherent in the communication network. We analyse real-world temporal networks and find that inhomogeneities most often dramatically slow down diffusion, hence the convergence process.

Figures

Figures reproduced from arXiv: 2607.03171 by Arash Badie-Modiri, Chiara Boldrini, J\'anos Kert\'esz, Lorenzo Valerio, M\'arton Karsai.

Figure 1
Figure 1. Figure 1: The composition of the parameters of each model under simple average [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Prediction accuracy for the parameters of each node after a perturba [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Relaxation of the diffusion process on random network models, [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Randomising structural and temporal heterogeneities increases the [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Influence of the source node (node 0) in the parameters of four nodes [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Residuals for parameters compared to diffusion process predictions. [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗

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

69 extracted references · 1 canonical work pages

  1. [1]

    Decentralized federated learning: Fundamentals, state of the art, frameworks, trends, and challenges,

    E. T. M. Beltrán, M. Q. Pérez, P. M. S. Sánchez, S. L. Bernal, G. Bovet, M. G. Pérez, G. M. Pérez, and A. H. Celdrán, “Decentralized federated learning: Fundamentals, state of the art, frameworks, trends, and challenges,”IEEE Communications Surveys & Tutorials, vol. 25, no. 4, pp. 2983–3013, 2023

    2023

  2. [2]

    Deep-fel: Decentralized, efficient and privacy-enhanced federated edge learning for healthcare cyber physical systems,

    Z. Lian, Q. Yang, W. Wang, Q. Zeng, M. Alazab, H. Zhao, and C. Su, “Deep-fel: Decentralized, efficient and privacy-enhanced federated edge learning for healthcare cyber physical systems,”IEEE Transactions on Network Science and Engineering, vol. 9, no. 5, pp. 3558–3569, 2022

    2022

  3. [3]

    Efficient ring- topology decentralized federated learning with deep generative models for medical data in ehealthcare systems,

    Z. Wang, Y . Hu, S. Yan, Z. Wang, R. Hou, and C. Wu, “Efficient ring- topology decentralized federated learning with deep generative models for medical data in ehealthcare systems,”Electronics, vol. 11, no. 10, p. 1548, 2022

    2022

  4. [4]

    Accelerating decentralized federated learning in heterogeneous edge computing,

    L. Wang, Y . Xu, H. Xu, M. Chen, and L. Huang, “Accelerating decentralized federated learning in heterogeneous edge computing,” IEEE Transactions on Mobile Computing, vol. 22, no. 9, pp. 5001–5016, 2022

    2022

  5. [5]

    Bdfl: A byzantine-fault-tolerance decentralized federated learning method for au- tonomous vehicle,

    J.-H. Chen, M.-R. Chen, G.-Q. Zeng, and J.-S. Weng, “Bdfl: A byzantine-fault-tolerance decentralized federated learning method for au- tonomous vehicle,”IEEE Transactions on Vehicular Technology, vol. 70, no. 9, pp. 8639–8652, 2021

    2021

  6. [6]

    Can decentralized algorithms outperform centralized algorithms? A case study for decentralized parallel stochastic gradient descent,

    X. Lian, C. Zhang, H. Zhang, C.-J. Hsieh, W. Zhang, and J. Liu, “Can decentralized algorithms outperform centralized algorithms? A case study for decentralized parallel stochastic gradient descent,” in Advances in Neural Information Processing Systems (NeurIPS), 2017. [Online]. Available: https://arxiv.org/abs/1705.09056

    Pith/arXiv arXiv 2017

  7. [7]

    𝑑 2: Decentralized training over decentralized data,

    H. Tang, X. Lian, M. Yan, C. Zhang, and J. Liu, “𝑑 2: Decentralized training over decentralized data,” inProceedings of the 35th International Conference on Machine Learning (ICML), ser. Proceedings of Machine Learning Research, vol. 80, 2018, pp. 4855–4863. [Online]. Available: https://arxiv.org/abs/1803.07068

    Pith/arXiv arXiv 2018

  8. [8]

    Robustness of decentralised learning to nodes and data disruption,

    L. Palmieri, C. Boldrini, L. Valerio, A. Passarella, M. Conti, and J. Kertész, “Robustness of decentralised learning to nodes and data disruption,”Computer Communications, p. 108250, 2025

    2025

  9. [9]

    Impact of network topology on the performance of decentralized federated learning,

    L. Palmieri, C. Boldrini, L. Valerio, A. Passarella, and M. Conti, “Impact of network topology on the performance of decentralized federated learning,”Computer Networks, vol. 253, p. 110681, 2024

    2024

  10. [10]

    Decentralized federated learning with distributed aggregation weight optimization,

    Z. Zhai, X. Yuan, X. Wang, and G. Y . Li, “Decentralized federated learning with distributed aggregation weight optimization,” 2025. [Online]. Available: https://arxiv.org/abs/2511.03284

    arXiv 2025

  11. [11]

    Initialisation and topology effects in decentralised federated learning,

    A. Badie-Modiri, C. Boldrini, L. Valerio, J. Kertész, and M. Karsai, “Initialisation and topology effects in decentralised federated learning,” Applied Network Science, vol. 10, p. 53, 2025

    2025

  12. [12]

    Barrat, M

    A. Barrat, M. Barthélemy, and A. Vespignani,Dynamical Processes on Complex Networks. Cambridge UP, 2012

    2012

  13. [13]

    Randomized reference models for temporal networks,

    L. Gauvin, M. Génois, M. Karsai, M. Kivelä, T. Takaguchi, E. Valdano, and C. L. Vestergaard, “Randomized reference models for temporal networks,”Siam Review, vol. 64, no. 4, pp. 763–830, 2022

    2022

  14. [14]

    Temporal networks,

    P. Holme and J. Saramäki, “Temporal networks,”Physics reports, vol. 519, no. 3, pp. 97–125, 2012

    2012

  15. [15]

    Time varying networks and the weakness of strong ties,

    M. Karsai, N. Perra, and A. Vespignani, “Time varying networks and the weakness of strong ties,”Scientific reports, vol. 4, no. 1, p. 4001, 2014

    2014

  16. [16]

    Small but slow world: How network topology and burstiness slow down spreading,

    M. Karsai, M. Kivelä, R. K. Pan, K. Kaski, J. Kertész, A.-L. Barabási, and J. Saramäki, “Small but slow world: How network topology and burstiness slow down spreading,”Physical Review E, vol. 83, no. 2, p. 025102, 2011

    2011

  17. [17]

    Random walks on temporal networks,

    M. Starnini, A. Baronchelli, A. Barrat, and R. Pastor-Satorras, “Random walks on temporal networks,”Physical Review E—Statistical, Nonlinear, and Soft Matter Physics, vol. 85, no. 5, p. 056115, 2012

    2012

  18. [18]

    Karsai, H.-H

    M. Karsai, H.-H. Jo, K. Kaskiet al.,Bursty human dynamics. Springer, 2018, vol. 1

    2018

  19. [19]

    The origin of bursts and heavy tails in human dynam- ics,

    A.-L. Barabasi, “The origin of bursts and heavy tails in human dynam- ics,”Nature, vol. 435, no. 7039, pp. 207–211, 2005. 12

    2005

  20. [20]

    Temporal networks: slowing down diffusion by long lasting interactions,

    N. Masuda, K. Klemm, and V . M. Eguíluz, “Temporal networks: slowing down diffusion by long lasting interactions,”Physical Review Letters, vol. 111, no. 18, p. 188701, 2013

    2013

  21. [21]

    Emergence of scaling in random net- works,

    A.-L. Barabási and R. Albert, “Emergence of scaling in random net- works,”science, vol. 286, no. 5439, pp. 509–512, 1999

    1999

  22. [22]

    Random graphs with arbitrary degree distributions and their applications,

    M. E. Newman, S. H. Strogatz, and D. J. Watts, “Random graphs with arbitrary degree distributions and their applications,”Physical review E, vol. 64, no. 2, p. 026118, 2001

    2001

  23. [23]

    Structure of growing networks with preferential linking,

    S. N. Dorogovtsev, J. F. F. Mendes, and A. N. Samukhin, “Structure of growing networks with preferential linking,”Physical review letters, vol. 85, no. 21, p. 4633, 2000

    2000

  24. [24]

    Collective dynamics of “small-world

    D. J. Watts and S. H. Strogatz, “Collective dynamics of “small-world” networks,”nature, vol. 393, no. 6684, pp. 440–442, 1998

    1998

  25. [25]

    The small world problem,

    S. Milgramet al., “The small world problem,”Psychology today, vol. 2, no. 1, pp. 60–67, 1967

    1967

  26. [26]

    Spatial networks,

    M. Barthélemy, “Spatial networks,”Physics reports, vol. 499, no. 1-3, pp. 1–101, 2011

    2011

  27. [27]

    Network motifs: simple building blocks of complex networks,

    R. Milo, S. Shen-Orr, S. Itzkovitz, N. Kashtan, D. Chklovskii, and U. Alon, “Network motifs: simple building blocks of complex networks,” Science, vol. 298, no. 5594, pp. 824–827, 2002

    2002

  28. [28]

    Network motifs: theory and experimental approaches,

    U. Alon, “Network motifs: theory and experimental approaches,”Nature Reviews Genetics, vol. 8, no. 6, pp. 450–461, 2007

    2007

  29. [29]

    Detecting community structure in networks,

    M. E. Newman, “Detecting community structure in networks,”The European physical journal B, vol. 38, no. 2, pp. 321–330, 2004

    2004

  30. [30]

    Community structure in social and biological networks,

    M. Girvan and M. E. Newman, “Community structure in social and biological networks,”Proceedings of the national academy of sciences, vol. 99, no. 12, pp. 7821–7826, 2002

    2002

  31. [31]

    Temporal motifs in time-dependent networks,

    L. Kovanen, M. Karsai, K. Kaski, J. Kertész, and J. Saramäki, “Temporal motifs in time-dependent networks,”Journal of Statistical Mechanics: Theory and Experiment, vol. 2011, no. 11, p. P11005, 2011

    2011

  32. [32]

    Decentralized stochastic optimization and gossip algorithms with compressed communication,

    A. Koloskova, S. U. Stich, and M. Jaggi, “Decentralized stochastic optimization and gossip algorithms with compressed communication,” in Proceedings of the 36th International Conference on Machine Learning (ICML), ser. Proceedings of Machine Learning Research, vol. 97, 2019, pp. 3478–3487. [Online]. Available: https://arxiv.org/abs/1902.00340

    Pith/arXiv arXiv 2019

  33. [33]

    Coordination-free decentralised federated learning on complex networks: Overcoming heterogeneity,

    L. Valerio, C. Boldrini, A. Passarella, J. Kertész, M. Karsai, and G. Iñiguez, “Coordination-free decentralised federated learning on complex networks: Overcoming heterogeneity,”arXiv preprint arXiv:2312.04504, 2023

    Pith/arXiv arXiv 2023

  34. [34]

    Consensus and coop- eration in networked multi-agent systems,

    R. Olfati-Saber, J. A. Fax, and R. M. Murray, “Consensus and coop- eration in networked multi-agent systems,”Proceedings of the IEEE, vol. 95, no. 1, pp. 215–233, 2007

    2007

  35. [35]

    Randomized gossip algorithms,

    S. Boyd, A. Ghosh, B. Prabhakar, and D. Shah, “Randomized gossip algorithms,”IEEE Transactions on Information Theory, vol. 52, no. 6, pp. 2508–2530, 2006. [Online]. Available: https: //web.stanford.edu/~boyd/papers/pdf/gossip.pdf

    2006

  36. [36]

    Coordination of groups of mobile autonomous agents using nearest neighbor rules,

    A. Jadbabaie, J. Lin, and A. S. Morse, “Coordination of groups of mobile autonomous agents using nearest neighbor rules,”IEEE Transactions on Automatic Control, vol. 48, no. 6, pp. 988–1001, 2003

    2003

  37. [37]

    Stability of multiagent systems with time-dependent com- munication links,

    L. Moreau, “Stability of multiagent systems with time-dependent com- munication links,”IEEE Transactions on Automatic Control, vol. 50, no. 2, pp. 169–182, 2005

    2005

  38. [38]

    Distributed subgradient methods for multi- agent optimization,

    A. Nedi ´c and A. Ozdaglar, “Distributed subgradient methods for multi- agent optimization,”IEEE Transactions on Automatic Control, vol. 54, no. 1, pp. 48–61, 2009

    2009

  39. [39]

    Decentralized federated learning: A segmented gossip approach,

    C. Hu, J. Jiang, and Z. Wang, “Decentralized federated learning: A segmented gossip approach,” 2019. [Online]. Available: https: //arxiv.org/abs/1908.07782

    Pith/arXiv arXiv 2019

  40. [40]

    Gossip learning as a decen- tralized alternative to federated learning,

    I. Heged˝ us, G. Danner, and M. Jelasity, “Gossip learning as a decen- tralized alternative to federated learning,” inDistributed Applications and Interoperable Systems (DAIS 2019), ser. Lecture Notes in Computer Science, vol. 11534. Springer, Cham, 2019, pp. 74–90

    2019

  41. [41]

    Privacy-preserving decentralized federated learning over time-varying communication graph,

    Y . Lu, Z. Yu, and N. Suri, “Privacy-preserving decentralized federated learning over time-varying communication graph,” 2022. [Online]. Available: https://arxiv.org/abs/2210.00325

    Pith/arXiv arXiv 2022

  42. [42]

    Asynchronous decentralized learning over unreliable wireless networks,

    E. Jeong, M. Zecchin, and M. Kountouris, “Asynchronous decentralized learning over unreliable wireless networks,” 2022. [Online]. Available: https://arxiv.org/abs/2202.00955

    Pith/arXiv arXiv 2022

  43. [43]

    Decentralized federated learning with gradient tracking over time-varying directed networks,

    D. T. A. Nguyen, S. Wang, D. T. Nguyen, A. Nedi ´c, and H. V . Poor, “Decentralized federated learning with gradient tracking over time-varying directed networks,” 2024. [Online]. Available: https://arxiv.org/abs/2409.17189

    Pith/arXiv arXiv 2024

  44. [44]

    Time-varying mixing matrix design for energy-efficient decentralized federated learning,

    X. Zhang, T. Nguyen, and T. He, “Time-varying mixing matrix design for energy-efficient decentralized federated learning,” 2025, conference version announced for IEEE INFOCOM 2026. [Online]. Available: https://arxiv.org/abs/2512.24069

    arXiv 2025

  45. [45]

    Decentralized federated learning over time-varying and heterogeneous mobile computing networks,

    B. Li, W. Gao, X. Deng, J. Xie, Z. Xiong, M. Siew, B. Guo, S. Mao, and Z. Han, “Decentralized federated learning over time-varying and heterogeneous mobile computing networks,”IEEE Transactions on Mobile Computing, 2025, early Access / publication details may vary by issue assignment

    2025

  46. [46]

    Anderson transitions,

    F. Evers and A. D. Mirlin, “Anderson transitions,”Reviews of Modern Physics, vol. 80, no. 4, pp. 1355–1417, 2008

    2008

  47. [47]

    Quantum google in a complex network,

    G. D. Paparo, M. Müller, F. Comellas, and M. A. Martin-Delgado, “Quantum google in a complex network,”Scientific reports, vol. 3, no. 1, p. 2773, 2013

    2013

  48. [48]

    Mixing times are hitting times of large sets,

    Y . Peres and P. Sousi, “Mixing times are hitting times of large sets,” Journal of Theoretical Probability, vol. 28, no. 2, pp. 488–519, 2015

    2015

  49. [49]

    Connectivity in wireless ad-hoc networks with a log-normal radio model,

    R. Hekmat and P. Van Mieghem, “Connectivity in wireless ad-hoc networks with a log-normal radio model,”Mobile networks and ap- plications, vol. 11, no. 3, pp. 351–360, 2006

    2006

  50. [50]

    The capacity of wireless networks,

    P. Gupta and P. R. Kumar, “The capacity of wireless networks,”IEEE Transactions on information theory, vol. 46, no. 2, pp. 388–404, 2002

    2002

  51. [51]

    Large-scale topo- logical and dynamical properties of the internet,

    A. Vázquez, R. Pastor-Satorras, and A. Vespignani, “Large-scale topo- logical and dynamical properties of the internet,”Physical Review E, vol. 65, no. 6, p. 066130, 2002

    2002

  52. [52]

    Topological patterns in street networks of self-organized urban settlements,

    C. Buhl, J. Gautrais, N. Reeves, R. V . Solé, S. Valverde, P. Kuntz, and G. Theraulaz, “Topological patterns in street networks of self-organized urban settlements,”The European Physical Journal B-Condensed Matter and Complex Systems, vol. 49, no. 4, pp. 513–522, 2006

    2006

  53. [53]

    Geographical dispersal of mobile communication networks,

    R. Lambiotte, V . D. Blondel, C. De Kerchove, E. Huens, C. Prieur, Z. Smoreda, and P. Van Dooren, “Geographical dispersal of mobile communication networks,”Physica A: Statistical Mechanics and its Applications, vol. 387, no. 21, pp. 5317–5325, 2008

    2008

  54. [54]

    Directed percolation in random temporal network models with heterogeneities,

    A. Badie-Modiri, A. K. Rizi, M. Karsai, and M. Kivelä, “Directed percolation in random temporal network models with heterogeneities,” Physical Review E, vol. 105, no. 5, p. 054313, 2022

    2022

  55. [55]

    Network spreading from network dimension,

    J. M. Moore, M. Small, G. Yan, H. Yang, C. Gu, and H. Wang, “Network spreading from network dimension,”Physical Review Letters, vol. 132, no. 23, p. 237401, 2024

    2024

  56. [56]

    Directed percolation in temporal networks,

    A. Badie-Modiri, A. K. Rizi, M. Karsai, and M. Kivelä, “Directed percolation in temporal networks,”Physical Review Research, vol. 4, no. 2, p. L022047, 2022

    2022

  57. [57]

    Point spectra of some mutually exciting point pro- cesses,

    A. G. Hawkes, “Point spectra of some mutually exciting point pro- cesses,”Journal of the Royal Statistical Society Series B: Statistical Methodology, vol. 33, no. 3, pp. 438–443, 1971

    1971

  58. [58]

    P. J. Laub, Y . Lee, and T. Taimre,The elements of Hawkes processes. Springer, 2021

    2021

  59. [59]

    Contact patterns among high school students,

    J. Fournet and A. Barrat, “Contact patterns among high school students,” PloS one, vol. 9, no. 9, p. e107878, 2014

    2014

  60. [60]

    Crawdad epfl/mobility,

    M. Piorkowski, N. Sarafijanovic-Djukic, and M. Grossglauser, “Crawdad epfl/mobility,” 2022. [Online]. Available: https://dx.doi.org/10.15783/ C7J010

    2022

  61. [61]

    Crawdad kth/campus,

    L. Pajevic, G. Karlsson, and V . Fodor, “Crawdad kth/campus,” 2022. [Online]. Available: https://dx.doi.org/10.15783/c7-5r6x-4b46

    work page doi:10.15783/c7-5r6x-4b46 2022

  62. [62]

    A poissonian explanation for heavy tails in e-mail communication,

    R. D. Malmgren, D. B. Stouffer, A. E. Motter, and L. A. Amaral, “A poissonian explanation for heavy tails in e-mail communication,” Proceedings of the National Academy of Sciences, vol. 105, no. 47, pp. 18 153–18 158, 2008

    2008

  63. [63]

    Dynamics of person-to-person interactions from distributed rfid sensor networks,

    C. Cattuto, W. Van den Broeck, A. Barrat, V . Colizza, J.-F. Pinton, and A. Vespignani, “Dynamics of person-to-person interactions from distributed rfid sensor networks,”PloS one, vol. 5, no. 7, p. e11596, 2010

    2010

  64. [64]

    Universal features of correlated bursty behaviour,

    M. Karsai, K. Kaski, A.-L. Barabási, and J. Kertész, “Universal features of correlated bursty behaviour,”Scientific reports, vol. 2, no. 1, p. 397, 2012

    2012

  65. [65]

    The role of the airline transportation network in the prediction and predictability of global epidemics,

    V . Colizza, A. Barrat, M. Barthélemy, and A. Vespignani, “The role of the airline transportation network in the prediction and predictability of global epidemics,”Proceedings of the National Academy of Sciences, vol. 103, no. 7, pp. 2015–2020, 2006

    2015

  66. [66]

    Epidemic spreading in scale-free networks,

    R. Pastor-Satorras and A. Vespignani, “Epidemic spreading in scale-free networks,”Physical review letters, vol. 86, no. 14, p. 3200, 2001

    2001

  67. [67]

    Wasserman and K

    S. Wasserman and K. Faust,Social network analysis: Methods and applications. Cambridge university press, 1994

    1994

  68. [68]

    The strength of weak ties,

    M. S. Granovetter, “The strength of weak ties,”American journal of sociology, vol. 78, no. 6, pp. 1360–1380, 1973

    1973

  69. [69]

    Error and attack tolerance of public transportation networks: a temporal networks approach,

    A. Badie-Modiri, “Error and attack tolerance of public transportation networks: a temporal networks approach,” Master’s thesis, Aalto Uni- versity, 2018. 13 APPENDIXA PERTURBATION EXPERIMENTS In this appendix we validate the diffusive relaxation ap- proximation for the propagation of a single-node perturbation under the decentralised pairwise-averaging dy...

    2018