Pith. sign in

REVIEW 3 major objections 7 minor 32 references

Small-Scale-Fading-Aware Resource Allocation in Wireless Federated Learning

T0 review · 3 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A multi-agent reinforcement learning policy that reallocates spectrum and power on every fast-fading time slot brings wireless federated learning close to the perfect-communication accuracy ceiling.

desk verdict Per-slot small-scale-fading-aware allocation is a solid idea, but the convergence theorem that anchors the reward is unproven as written and the reward-to-bound link is heuristic. read the letter →

arxiv 2505.03533 v1 pith:3Z3IALCF submitted 2025-05-06 cs.LG

Jiacheng Wang , Le Liang , Hao Ye , Chongtao Guo , Shi Jin This is my paper · ORCID
classification cs.LG
keywords federatedlearningsmall-scalefadingresourceallocationmulti-agentreinforcementQMIXDec-POMDPconvergenceanalysiswirelesscommunications
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

Wireless federated learning usually assumes the channel is fixed during a round of gradient uploads, but real uploads span many fast-fading coherence slots. This paper claims that letting each client re-decide its sub-band and transmit power at every coherence slot, coordinated by the multi-agent QMIX algorithm, improves final model accuracy when data are non-IID and channels fluctuate. The argument rests on a one-step convergence bound for FedAvg whose only round-controllable term is the gap between the ideally-uploaded aggregate gradient and the gradient the server actually receives; the paper converts that gap into a per-slot reward. In simulations the learned policy beats heuristic baselines and approaches the perfect-communication upper bound, and ablations show both the small-scale fading observation and a gradient-deviation observation are needed for the gain.

What carries the argument

The central object is the convergence-reward link: Theorem 1 isolates $C_2\mathbb{E}\|\bar{\tilde g}_t-\tilde g_t\|^2$, the mean squared error between the aggregate stochastic gradient that would be uploaded if every client succeeded and the gradient the server actually aggregates, and this term depends on which clients the wireless schedule lets through. Because the true ideal gradient is unavailable, the paper approximates it with the previous round's aggregate gradient $\tilde g_{t-1}$, which clients can recover from the difference of the last two broadcast global models, yielding per-client estimated deviations $\tilde\delta_{n,t}^2$. The reward then decomposes that expression over time slots as a client-count term plus gradient-deviation and pairwise inner-product terms, together with a sum-rate term. QMIX supplies centralized training via a monotonic mixing network and decentralized execution, so each client's DQN maps its local observation to sub-band and power actions per slot.

What would settle it

Train the same QMIX setup with the reward computed from the true ideally-uploaded gradient $\bar{\tilde g}_t$, available at the server during centralized training, instead of the previous-round proxy $\tilde g_{t-1}$; if accuracy does not improve, or if removing the convergence-reward terms leaves performance unchanged, the reported gain is not caused by the convergence-derived reward.

Watch

Extended reading notes

Core claim

The paper claims that resource allocation for wireless FedAvg should be decided below the round level: inside one gradient-upload round of $T_s$ coherence-time slots, each client can pick a sub-band and a power level per slot, and a QMIX-trained multi-agent policy can learn to make those choices well. The enabling analysis is Theorem 1, a one-step convergence bound of the form $\mathbb{E}[F(w_{t+1})]-F(w_t)\le C_1\|\nabla F(w_t)\|^2+C_2\mathbb{E}\|\bar{\tilde g}_t-\tilde g_t\|^2+C_3$, where the middle term is the only component that the current round's upload decisions can shrink. The paper therefore builds a Dec-POMDP in which each client observes large-scale fading, per-slot small-scale fading across sub-bands, remaining transmission volume, an estimated gradient deviation $\tilde\delta_{n,t}^2=\|\tilde g_{t-1}-\tilde g_{n,t}\|^2$, and the round index, and receives a reward that heuristically splits the convergence term into per-slot pieces plus a throughput incentive. On CIFAR-10 with time-varying multipath channels generated from an urban microcell model, the resulting policy consistently beats max-sum-rate, max-individual-rate, and random baselines, and tracks the perfect-communication accuracy ceiling, most clearly under strong statistical heterogeneity ($\alpha=0.5$).

Load-bearing premise

The scheme depends on the assumption that replacing the true ideally-uploaded gradient with the previous round's aggregate gradient, and then decomposing the convergence term into the heuristic per-slot reward of Eqs. (29)-(31), preserves which policies are best under the actual objective; if that proxy is poor, the learned allocation optimizes a reward only loosely connected to convergence.

Editorial extensions

If this is right

  • Per-slot, per-client spectrum and power choices can shrink the round-level aggregation-error term in the convergence bound, something a static per-round allocation cannot do.
  • Under strong statistical heterogeneity, steering resources toward clients whose gradients align with the aggregate matters more than maximizing total throughput, which is why the QMIX policy outperforms sum-rate-maximizing baselines.
  • A policy trained at one heterogeneity level and channel condition remains effective when those settings change at test time, since the paper fixes $\alpha=0.5$, $n_c=21$ during training and varies both during testing.
  • The ablation results imply that both the small-scale fading observation and the gradient-deviation observation carry information the agent needs; removing either one lowers accuracy.

Reading between the lines

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

  • A stress test the paper does not run is to compare the proxy reward against an oracle reward computed from the true ideally-uploaded gradient at the server; if the policy ordering changes, the reported gain may be driven by the throughput term rather than the convergence term.
  • The framework suggests an immediate extension to explicit client selection at the start of each round, named as future work in the conclusion, and the same reward structure could be adapted to FL variants such as SCAFFOLD whose drift terms enter the bound differently.
  • Because the proxy approximates the current ideal gradient with the previous round's aggregate, abrupt shifts in data distribution between rounds are a plausible failure mode worth testing.
  • The decentralized-execution phase requires every agent to know the large-scale fading of all clients, an overhead that grows with the number of clients and is not quantified in the paper.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. This paper considers spectrum and power allocation for the uplink of wireless federated learning (FL) when small-scale fading varies within a single gradient-upload round, so the block-fading assumption in prior work is inadequate. The authors derive a one-step convergence bound for FedAvg under non-IID data (Theorem 1), whose second term depends on the set of clients that successfully upload, and use this term to design a Dec-POMDP: each client is an agent that chooses a sub-band and a power level in every coherence-time slot, with the problem solved by QMIX under centralized training and decentralized execution. Agent observations include large- and small-scale fading, transmission progress, and a gradient-deviation term defined as the distance between the client's current local gradient and the previous round's aggregate gradient. The reward combines a convergence-related component with a transmission-rate component. Experiments on CIFAR-10 with QuaDRiGa channels compare the proposed scheme with three heuristic baselines over nine (statistical-heterogeneity, channel-fluctuation) settings; the QMIX policy consistently attains the highest test accuracy and approaches a perfect-communication upper bound, and an ablation (Fig. 6) shows that removing either the small-scale-fading or the gradient-deviation observation degrades performance.

Significance. If the results hold, this is a solid systems contribution to wireless FL: the per-slot, small-scale-fading-aware allocation framework is a practical improvement over round-static schedulers, and the empirical gains over the three baselines are clear and consistent across heterogeneity levels. The paper's strongest assets are the well-motivated problem, the realistic QuaDRiGa evaluation, the breadth of tested scenarios, and the informative ablation of Fig. 6. The theoretical anchor (Theorem 1) is not proven as written, and the reward's link to the convergence bound is heuristic in three unquantified respects (the g-tilde_{t-1} proxy, the dropped 1/|N_t|^2 factor, and the non-telescoping per-slot decomposition). Because the paper's narrative attributes the gains to this convergence-informed design, the theory-reward chain needs to be repaired and validated before the attribution claim is credible. The errors appear fixable within the scope of a revision.

major comments (3)
  1. [Appendix C (Theorem 1); Appendix A (Lemma 1)] The proof of Theorem 1 is not correct as written, although the final statement appears salvageable. (i) In the bounding of E||grad F(w_t) - g-tilde_t||^2, the paper uses the deterministic inequality ||a+b+c||^2 <= 2||a||^2+2||b||^2+||c||^2 (implicit in the step bounding by 2(A1) + 2(A2) + (A3)); this inequality is false, e.g., a=(1,0), b=(0,1), c=(1,1) gives 8 <= 6. The intended estimate can be recovered in expectation from the unbiased-sampling assumption because g-bar_t - g-bar-tilde_t has conditional mean zero, but that argument is not supplied. (ii) The expansion of E||grad F(w_t) - g-bar_t||^2 displays the identity (1/E) sum_i grad F_n(w_t) - eta_l sum_i grad F_n(w_{n,t}^{(i-1)}) = (1 - eta_l E) sum_i grad F_n(w_t) - eta_l E (sum_i grad F_n(w_t) - sum_i grad F_n(w_{n,t}^{(i-1)})); expanding the right-hand side gives (1 - 2 eta_l E) sum_i grad F_n(w_t) + eta_l E sum_i grad F_n(w_{n,t}^{(i-1)}), which differs from the left-hand side, and the sign of the second term should be positive. The following lines (46)-(50) are consistent with the corrected identity, so the error is repairable, but the appendix must be rewritten. (iii) In Appendix A, the step between (38) and (39) asserts that eta_l <= 1/sqrt(8 E L) implies 1 + 1/(E-1) >= 1 + 1/(2E-1) + 4 E eta_l^2 L^2; this fails for E >= 3 (for E=3 and eta_l^2 L^2 = 1/24, the inequality reads 0.5 >= 0.7), and a sufficient condition would instead be eta_l^2 L^2 <= 1/(4(E-1)(2E-1)).
  2. [Section IV-D, Eqs. (28)-(31)] The reward is a heuristic surrogate whose connection to the quantity Theorem 1 identifies (E||g-bar-tilde_t - g-tilde_t||^2) is unvalidated at three levels. First, g-bar-tilde_t is replaced by g-tilde_{t-1} (the previous round's aggregate over the then-successful subset) in the observation (26) and in (28)-(29), with no bound on ||g-bar-tilde_t - g-tilde_{t-1}||; the global model changes each round, so this approximation error is uncontrolled, particularly in early FL rounds. Second, the reward in (29) drops the 1/|N_t|^2 normalization present in (28) and adds lambda_1 |N_t|, so the reward landscape differs from the convergence term in a way that can change the optimal policy (e.g., favoring a small unnormalized deviation ||sum_{n in N_t} delta-tilde_n||^2 over a large |N_t|). Third, the per-slot reward (30) does not decompose (29): since N_ts is cumulative, applying (30) at every slot after the first success counts each client's terms repeatedly, and the factors 2 and 1 on the cross terms do not reproduce the pairwise expansion of (29). Because Section V.B attributes the QMIX gains to this convergence-derived reward and the ablation in Fig. 6 attributes them to the gradient-deviation observation, the paper should provide direct evidence that the proxy preserves policy ordering, e.g., by reporting the realized true term E||g-bar-tilde_t - g-tilde_t||^2 (computable in simulation) under the learned policy and baselines, or by ablating the reward components (with/without the proxy, with/without the 1/|N_t|^2 factor). Without such evidence, the reported gains could equally stem from the lambda_t sum-rate term and from QMIX coordination alone.
  3. [Section V-A, Tables I-II] The reward weights in the central experiments are not reported: lambda_1 and lambda_2 in (29)-(30) are defined but their values are absent from Tables I and II, and lambda_c and lambda_t in (31), described as balancing and normalizing the convergence and throughput rewards, are given no values and no normalization procedure. The two reward components have very different scales (gradient-norm squared terms versus bits-per-slot rates), so without these values the reward actually optimized in the experiments is unknown to the reader and the results are not reproducible. Given that the paper's central claim is that the gradient-deviation mechanism drives the improvement, the authors should report the chosen weights and ideally include a sensitivity study over them.
minor comments (7)
  1. [Section III-B] The statement that the first and third terms in (25) 'are determined only by the initialization and FL hyperparameters' is inaccurate: the first term contains ||grad F(w_t)||^2, which depends on the whole training trajectory and the data, and C3 contains the data-dependent constants sigma_g^2 and sigma_l^2. A wording like 'do not depend on the current upload decisions' would preserve the design rationale.
  2. [Appendix C, line preceding Eq. (50)] The intermediate line writes '24/N eta_l^4 E^4 L^2 sum_n 4E ||F_n(w_t)||^2', which carries a spurious factor 4E; the final constant in (50) is consistent with the factor-free expression, so this appears to be a typographical slip that should be fixed.
  3. [Appendix A, Eqs. (36)-(40)] The stochastic gradient in the local update is written as grad-tilde F_n(w_{n,t}^{(i)}) but should be evaluated at w_{n,t}^{(i-1)}; likewise the summation in (40) is unclear, since for i=1 the stated range p=1,...,i-1 is empty yet the bound should still be positive.
  4. [Section V-A] The sentence 'each agent interacts with the recurring environment 20 times per FL round' is ambiguous; it should be clarified whether these are environment steps, RL episodes, or network gradient updates, and how they relate to the T_s=250-slot upload process and to the global model update in Algorithm 1.
  5. [Fig. 4] The shaded bands are the 10th-90th percentile ranges of only 5 independent runs; the authors should state the percentile estimator used or show all runs, since percentile bands from 5 samples are noisy.
  6. [Throughout] Minor typos: 'gredients' appears twice in the text around Eq. (30) and should read 'gradients'; the caption of Fig. 6 says 'gradient derivation' where 'gradient deviation' is meant.
  7. [Section V-B] The baseline set contains only heuristic schemes; comparing against a state-of-the-art FL-aware scheduler from the block-fading literature (e.g., [16] or [17]) would strengthen the claim that the gains come specifically from exploiting small-scale fading within the round rather than from better coordination in general.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the convergence bound is proven independently, the reward is explicitly heuristic, and the evaluation uses external FL test accuracy rather than the bound itself.

full rationale

The paper's derivation chain is self-contained and non-circular. Theorem 1 is a one-step FedAvg convergence bound proven in Appendices A-C from Assumptions 1-3; it does not assume the MARL method or the reward. The reward in Eqs. (28)-(31) is derived from the second term of that bound, but the paper explicitly labels the decomposition as heuristic: 'To facilitate decomposition, we heuristically assign the convergence-related reward for the t-th FL round as' (Section IV-D). Moreover, the central empirical claim is evaluated by external FL test accuracy on CIFAR-10 against heuristic baselines and a perfect-communication upper bound (Section V-B, Fig. 4), not by the convergence bound itself. The ablation in Fig. 6 tests whether omitting gradient-deviation or small-scale-fading observations changes test accuracy, which is an external behavioral check rather than a restatement of the reward. The self-citations used in the paper (e.g., [21], [23] for spectrum-sharing schemes and [5] for unreliable-communication FL) provide modeling background and are not load-bearing for the main convergence or performance claims. The main weakness, namely that the heuristic reward may not preserve policy ordering under the true convergence objective, is a correctness or robustness risk, not a circularity: the paper's own text explicitly acknowledges the heuristic nature of the reward, and the empirical evaluation does not reduce to the optimized objective.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The theorem rests on standard FL assumptions plus two paper-specific heuristics: approximating the ideal gradient with the previous round's aggregate, and factorizing the convergence term into a per-slot reward. The reward weights lambda1, lambda2, lambdac, lambdat are free parameters whose values are not reported. No new physical entities are introduced.

free parameters (3)
  • Reward weight lambda1 = not reported
    Balances the bonus for successful uploads in Eq. (29); selected to ensure positive reward but value omitted.
  • Reward weight lambda2 = not reported
    Balances the penalty for gradient deviation and inner-product terms in Eq. (29); value and selection procedure omitted.
  • Reward combining weights lambdac and lambdat = not reported
    Balance convergence-related reward and transmission-rate reward in Eq. (31); values omitted.
assumptions (6)
  • domain assumption L-smoothness of local objectives (Assumption 1)
    Standard smoothness required for the descent argument in Theorem 1; not verified for the CIFAR-10 CNN.
  • domain assumption Unbiased mini-batch sampling and bounded local variance (Assumption 2)
    Standard stochastic-gradient assumption used in Lemma 1 and Theorem 1.
  • domain assumption Bounded global variance (Assumption 3)
    Quantifies non-IID heterogeneity; used to bound average squared gradient norms.
  • domain assumption Error-free downlink broadcast (Section II-A)
    Assumes global model download is reliable; not realistic for all wireless settings but common in literature.
  • ad hoc to paper The approximate identity between the ideal gradient and the previous round's aggregate gradient in observations and reward (Eqs. (26), (28)-(29))
    Needed to make gradient-deviation locally computable; no justification that previous-round aggregate represents current ideal gradient.
  • ad hoc to paper Heuristic per-slot reward factorization (Eqs. (29)-(31)) is a valid proxy for the convergence term
    The decomposition is motivated by Eq. (28) but is not equivalent to it and not independently validated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Small-Scale-Fading-Aware Resource Allocation in Wireless Federated Learning." pith.science (2026). https://pith.science/paper/3Z3IALCF

@misc{pith2026250503533,
  author       = {Pith},
  title        = {Pith review of: Small-Scale-Fading-Aware Resource Allocation in Wireless Federated Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3Z3IALCF}},
  note         = {Machine review of arXiv:2505.03533}
}
read the original abstract

Judicious resource allocation can effectively enhance federated learning (FL) training performance in wireless networks by addressing both system and statistical heterogeneity. However, existing strategies typically rely on block fading assumptions, which overlooks rapid channel fluctuations within each round of FL gradient uploading, leading to a degradation in FL training performance. Therefore, this paper proposes a small-scale-fading-aware resource allocation strategy using a multi-agent reinforcement learning (MARL) framework. Specifically, we establish a one-step convergence bound of the FL algorithm and formulate the resource allocation problem as a decentralized partially observable Markov decision process (Dec-POMDP), which is subsequently solved using the QMIX algorithm. In our framework, each client serves as an agent that dynamically determines spectrum and power allocations within each coherence time slot, based on local observations and a reward derived from the convergence analysis. The MARL setting reduces the dimensionality of the action space and facilitates decentralized decision-making, enhancing the scalability and practicality of the solution. Experimental results demonstrate that our QMIX-based resource allocation strategy significantly outperforms baseline methods across various degrees of statistical heterogeneity. Additionally, ablation studies validate the critical importance of incorporating small-scale fading dynamics, highlighting its role in optimizing FL performance.

Figures

Figures reproduced from arXiv: 2505.03533 by the authors.

Figure 2
Figure 2. QMIX network: The blue block on the right represents the agent [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 1
Figure 1. Workflow of the proposed resource allocation strategy: In the [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 3
Figure 3. The distribution of local training data across different degrees of statistical heterogeneity. Each column represents the dataset of an individual client, [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Test accuracy over the number of training rounds under different levels of system and statistical heterogeneity. The concentration parameter [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Cumulative global reward for each training episode with increasing [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 24 canonical work pages

  1. [1]

    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 Proc. Artif. Intell. Statist. (AIS- TATS), Apr. 2017, pp. 1273–1282

    work page 2017

  2. [2]

    Towards federated learning at scale: System design,

    K. Bonawitz, “Towards federated learning at scale: System design,” 2019, arXiv:1902.01046

    arXiv 2019

  3. [3]

    Federated learning and wireless communications,

    Z. Qin, G. Y . Li, and H. Ye, “Federated learning and wireless communications,” IEEE Wirel. Commun. , vol. 28, no. 5, pp. 134–140, Oct. 2021

    work page 2021

  4. [4]

    Multi-objective energy- efficient resource allocation for multi-RAT heterogeneous net- works,

    G. Yu, Y . Jiang, L. Xu, and G. Y . Li, “Multi-objective energy- efficient resource allocation for multi-RAT heterogeneous net- works,” IEEE J. Sel. Areas Commun., vol. 33, no. 10, pp. 2118– 2127, Oct. 2015

    work page 2015

  5. [5]

    Decentralized federated learning with unreliable communications,

    H. Ye, L. Liang, and G. Y . Li, “Decentralized federated learning with unreliable communications,” IEEE J. Sel. Topics Signal Process., vol. 16, no. 3, pp. 487–500, Apr. 2022

    work page 2022

  6. [6]

    On the convergence of FedAvg on non-IID data,

    X. Li, K. Huang, W. Yang, S. Wang, and Z. Zhang, “On the convergence of FedAvg on non-IID data,” in Proc. Int. Conf. Learn. Represent. (ICLR) , 2020, pp. 1–26

    work page 2020

  7. [7]

    Scaffold: Stochastic controlled averaging for federated learning,

    S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. Stich, and A. T. Suresh, “Scaffold: Stochastic controlled averaging for federated learning,” in Proc. Int. Conf. Mach. Learn. (ICML) , 2020, pp. 5132–5143

    work page 2020

  8. [8]

    Scheduling policies for federated learning in wireless networks,

    H. H. Yang, Z. Liu, T. Q. Quek, and H. V . Poor, “Scheduling policies for federated learning in wireless networks,” IEEE Trans. Commun., vol. 68, no. 1, pp. 317–333, Jan. 2020

    2020

Show all 32 references
  1. [9]

    Fed- erated learning with non-IID data,

    Y . Zhao, M. Li, L. Lai, N. Suda, D. Civin, and V . Chandra, “Fed- erated learning with non-IID data,” 2018, arXiv:1806.00582

    2018 arXiv

  2. [10]

    Measuring the effects of non-identical data distribution for federated visual classifica- tion,

    T.-M. H. Hsu, H. Qi, and M. Brown, “Measuring the effects of non-identical data distribution for federated visual classifica- tion,” 2019, arXiv:1909.06335

    2019 arXiv

  3. [11]

    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,” in Proc. IEEE Int. Conf. Commun. (ICC) , 2019, pp. 1–7

    2019

  4. [12]

    Federated learning over wireless networks: Optimization model design and analysis,

    N. H. Tran, W. Bao, A. Zomaya, M. N. Nguyen, and C. S. Hong, “Federated learning over wireless networks: Optimization model design and analysis,” in Proc. IEEE Conf. Comput. Commun. (IEEE INFOCOM) , Apr. 2019, pp. 1387–1395

    2019

  5. [13]

    Joint device scheduling and bandwidth allocation for federated learning over wireless networks,

    T. Zhang, K.-Y . Lam, J. Zhao, and J. Feng, “Joint device scheduling and bandwidth allocation for federated learning over wireless networks,” IEEE Trans. Wireless Commun., Jul. 2023

    2023

  6. [14]

    Optimizing federated learning on non-IID data with reinforcement learning,

    H. Wang, Z. Kaplan, D. Niu, and B. Li, “Optimizing federated learning on non-IID data with reinforcement learning,” in Proc. IEEE Conf. Comput. Commun. (IEEE INFOCOM) , Jul. 2020, pp. 1698–1707

    2020

  7. [15]

    FedQMIX: Communication-efficient feder- ated learning via multi-agent reinforcement learning,

    S. Cao, H. Zhang, T. Wen, H. Zhao, Q. Zheng, W. Zhang, and D. Zheng, “FedQMIX: Communication-efficient feder- ated learning via multi-agent reinforcement learning,” High- Confidence Comput., vol. 4, no. 2, p. 100179, Jun. 2024

    2024

  8. [16]

    Joint de- vice scheduling and resource allocation for latency constrained wireless federated learning,

    W. Shi, S. Zhou, Z. Niu, M. Jiang, and L. Geng, “Joint de- vice scheduling and resource allocation for latency constrained wireless federated learning,” IEEE Trans. Wireless Commun. , vol. 20, no. 1, pp. 453–467, Sep. 2020

    2020

  9. [17]

    A joint learning and communications framework for federated learning over wireless networks,

    M. Chen, Z. Yang, W. Saad, C. Yin, H. V . Poor, and S. Cui, “A joint learning and communications framework for federated learning over wireless networks,” IEEE Trans. Wireless Com- mun., vol. 20, no. 1, pp. 269–283, Oct. 2020

    2020

  10. [18]

    Joint client selection and bandwidth allocation of wireless federated learning by deep reinforcement learning,

    W. Mao, X. Lu, Y . Jiang, and H. Zheng, “Joint client selection and bandwidth allocation of wireless federated learning by deep reinforcement learning,” IEEE Trans. Services Comput., vol. 17, no. 1, pp. 336–348, Jan. 2024

    2024

  11. [19]

    A multi-agent reinforcement learning approach for efficient client selection in federated learning,

    S. Q. Zhang, J. Lin, and Q. Zhang, “A multi-agent reinforcement learning approach for efficient client selection in federated learning,” in Proc. AAAI Conf. Artif. Intell., vol. 36, no. 8, 2022, pp. 9091–9099

    2022

  12. [20]

    FedAEB: Deep reinforcement learning based joint client selection and resource allocation strategy for heterogeneous federated learning,

    F. Zheng, Y . Sun, and B. Ni, “FedAEB: Deep reinforcement learning based joint client selection and resource allocation strategy for heterogeneous federated learning,”IEEE Trans. Veh. Technol., vol. 73, no. 6, pp. 8835–8846, Jun. 2024

    2024

  13. [21]

    Spectrum sharing in vehicular networks based on multi-agent reinforcement learning,

    L. Liang, H. Ye, and G. Y . Li, “Spectrum sharing in vehicular networks based on multi-agent reinforcement learning,” IEEE J. Sel. Areas Commun. , vol. 37, no. 10, pp. 2282–2292, Oct. 2019

    2019

  14. [22]

    Experience-driven computational resource allocation of federated learning by deep reinforcement learning,

    Y . Zhan, P. Li, and S. Guo, “Experience-driven computational resource allocation of federated learning by deep reinforcement learning,” in Proc. IEEE Int. Parallel Distrib. Process. Symp. (IPDPS), Jul. 2020, pp. 234–243

    2020

  15. [23]

    Deep reinforcement learning based resource allocation for V2V communications,

    H. Ye, G. Y . Li, and B.-H. F. Juang, “Deep reinforcement learning based resource allocation for V2V communications,” IEEE Trans. Veh. Technol., vol. 68, no. 4, pp. 3163–3173, Apr. 2019

    2019

  16. [24]

    Federated learning in unreliable and resource-constrained cellular wireless networks,

    M. Salehi and E. Hossain, “Federated learning in unreliable and resource-constrained cellular wireless networks,” IEEE Trans. Commun., vol. 69, no. 8, pp. 5136–5151, Aug. 2021

    2021

  17. [25]

    Monotonic value function factorisation for deep multi-agent reinforcement learning,

    T. Rashid, M. Samvelyan, C. S. De Witt, G. Farquhar, J. Foer- ster, and S. Whiteson, “Monotonic value function factorisation for deep multi-agent reinforcement learning,” J. Mach. Learn. Res., vol. 21, no. 178, pp. 1–51, 2020

    2020

  18. [26]

    Analog gradient aggregation for federated learning over wireless networks: Customized design and convergence analysis,

    H. Guo, A. Liu, and V . K. Lau, “Analog gradient aggregation for federated learning over wireless networks: Customized design and convergence analysis,” IEEE Internet Things J. , vol. 8, no. 1, pp. 197–210, Jan. 2021

    2021

  19. [27]

    Stabilising experience replay for deep multi-agent reinforcement learning,

    J. Foerster, N. Nardelli, G. Farquhar, T. Afouras, P. H. Torr, P. Kohli, and S. Whiteson, “Stabilising experience replay for deep multi-agent reinforcement learning,” in Proc. Int. Conf. Mach. Learn. (ICML) , 2017, pp. 1146–1155

    2017

  20. [28]

    Learning multiple layers of features from tiny images,

    A. Krizhevsky, G. Hinton et al. , “Learning multiple layers of features from tiny images,” 2009

    2009

  21. [29]

    QuaDRiGa: A 3-D multi-cell channel model with time evo- lution for enabling virtual field trials,

    S. Jaeckel, L. Raschkowski, K. B ¨orner, and L. Thiele, “QuaDRiGa: A 3-D multi-cell channel model with time evo- lution for enabling virtual field trials,” IEEE Trans. Antennas Propag., vol. 62, no. 6, pp. 3242–3256, Mar. 2014

    2014

  22. [30]

    Federated learning based on dynamic regularization,

    D. A. E. Acar, Y . Zhao, R. M. Navarro, M. Mattina, P. N. Whatmough, and V . Saligrama, “Federated learning based on dynamic regularization,” 2021, arXiv:2111.04263

    2021 arXiv

  23. [31]

    Fedlab: A flexible federated learning framework,

    D. Zeng, S. Liang, X. Hu, H. Wang, and Z. Xu, “Fedlab: A flexible federated learning framework,” J. Mach. Learn. Res. , vol. 24, no. 100, pp. 1–7, 2023

    2023

  24. [32]

    An overview of gradient descent optimization algo- rithms,

    S. Ruder, “An overview of gradient descent optimization algo- rithms,” 2016, arXiv:1609.04747

    2016 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.