REVIEW 4 major objections 5 minor 1 cited by
Adaptive Quantization Resolution and Power Control for Federated Learning over Cell-free Networks
T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash
Pith's one-line read Federated learning over cell-free networks can cut uplink communication overhead by at least 93% using adaptive mixed-resolution quantization that spends full bits only on the largest gradient entries, while keeping test accuracy close to…
desk verdict The mixed-resolution quantizer is a sensible idea, but the bitstream omits a mask, so the headline overhead numbers are off by about a factor of two; still worth a careful revision. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the adaptive mixed-resolution quantization rule in (6)-(7) together with the rate-per-bit objective in (13). The quantizer exploits the empirically observed sparsity of local gradients: the fraction $s_j^t$ of entries above the magnitude ratio threshold $\lambda_j$ is typically below 2% in the experiments, so roughly 98% of entries are transmitted as a single sign bit. The power control solves the minimax latency problem by bisection over the lowest rate-per-bit ratio and a linear program for the user powers, explicitly balancing the number of bits each user owes against the rate the cell-free network can give that user.
What would settle it
Run the same algorithm on a model class whose gradients are known to be dense, such as a small ResNet or a transformer on a language task, and measure the average high-resolution fraction $s$ and the accuracy gap to full-precision FL; if $s$ exceeds roughly 5-10%, the claimed at-least-93% overhead reduction cannot hold, and the convergence bound shows the gap grows with $K c_{max}^2$.
Extended reading notes
Core claim
The central claim is that a threshold-based mixed-resolution quantizer defined by $\lambda_j$ relative to the infinity norm of each local gradient update preserves federated learning convergence while making uplink traffic almost negligible. For entries with $|[\delta w_j^t]_i| / \|\delta w_j^t\|_\infty < \lambda_j$, only the sign is sent; the few entries at or above the threshold are uniformly quantized with $b_j$ bits, and the reconstruction sets the low-resolution magnitudes to half the quantized smallest high-resolution value. Coupled with a min-max power control that maximizes the smallest rate-per-bit ratio via bisection and linear programming, the method lets more FL iterations fit in a given latency budget, which is why it gains 10% accuracy over fixed-quantization benchmarks under the same total time.
Load-bearing premise
The entire overhead reduction rests on the empirical regularity that most local gradient entries are so close to zero that replacing their magnitude with one sign bit does not meaningfully hurt convergence; the paper measures this sparsity ($s \approx 0.7\%\text{--}1.8\%$) on three datasets but gives no theoretical guarantee or cross-architecture evidence that it always holds.
Editorial extensions
If this is right
- Uplink communication per FL round no longer scales with 32 bits times model dimension; with $s \approx 1\%$, the per-user payload is dominated by one bit per parameter, reducing overhead by at least 93% on the tested datasets.
- Because the bit count $b_j^t$ adapts per user and iteration, the power control can equalize finish times, so iteration latency is set by the maximum rate-per-bit rather than the worst channel user.
- Under a fixed total latency budget, the method completes more global rounds (up to $T_{max}=27$ versus 16 for AQUILA in the reported setting), which is the direct mechanism for the 10% accuracy gain over benchmarks.
- The convergence bound (10) shows the quantization error enters only through $K c_{max}^2(2L + T \bar\sigma^2)$, so as long as the per-entry error bound $c_j$ stays small, AdaGrad FL converges at the same order as unquantized adaptive federated optimization.
Reading between the lines
- The sparsity assumption is the hinge: the reported 93-96% savings reflect $s \approx 0.7\%\text{--}1.8\%$ on these CNN/dataset combinations. For architectures with denser gradients (residual networks, transformers, or batch-normalized models), the savings would shrink and the accuracy gap could widen; that is a testable extension, not a claim in the paper.
- The low-resolution entries are reconstructed at half the smallest high-resolution quantized value, which injects a systematic magnitude bias toward zero. A bias-correction or dithering variant could push accuracy closer to full precision while keeping the same bit budget.
- The per-user threshold $\lambda_j$ is fixed in the experiments; adapting it to channel quality (weaker users could send fewer high-resolution entries) would trade convergence per round against latency exactly along the rate-per-bit curve the paper already defines.
- The same mixed-resolution idea could be applied to the downlink broadcast of the global model, where the paper assumes error-free transmission; the bit-saving logic does not depend on the uplink-only setting.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes an adaptive mixed-resolution quantization scheme for federated learning over cell-free massive MIMO: gradient entries below a magnitude-ratio threshold are encoded with a single sign bit, while the remaining entries are uniformly quantized with b_j bits, and uplink powers are optimized by bisection over an epigraph-form min-max problem to mitigate stragglers. The authors claim a convergence bound for FL with local AdaGrad updates (Proposition 1), overhead reductions of about 93–96% relative to full-precision FL with comparable accuracy on CIFAR-10, CIFAR-100, and Fashion-MNIST, and a 75% overhead reduction and 10% accuracy gain over AQUILA, LAQ, and Top-q under a fixed latency budget (Tables II–III).
Significance. The problem is timely and the system model is detailed; the power-control reformulation in Eq. (14) is elegant, and the bisection/LP solution is a sensible approach. The paper also makes a concrete attempt to prove a quantization-error bound in Lemma 1. If the central claims held, the work would be a useful contribution to communication-efficient FL over cell-free networks. However, the protocol as written is not implementable because the mixed-resolution bitstream is not uniquely decodable, and the main theoretical bound has a serious proof gap. Both issues are load-bearing for the headline overhead and accuracy claims, so the manuscript needs substantial revision.
major comments (4)
- [Section II-C, Eq. (7)–(8)] The serialized bitstream is not uniquely decodable. Low-resolution entries are coded as a single bit (0 or 1), high-resolution entries as a b_j-bit uniform-quantizer codeword, and the total is counted as d((1-s)+b_j s)+32 bits with no mask, index list, or prefix code. There is no specified restriction preventing a one-bit codeword from being a prefix of a b_j-bit codeword, so a receiver cannot determine which coordinates use which length. If a d-bit mask is added, the true per-vector cost becomes d(2+(b_j-1)s)+32 bits; for the Table III configuration (b_j=4, s=0.044%) this changes the claimed overhead reduction versus LAQ/AQUILA from about 75% to about 50%, and T_max and the reported accuracy margin must be recomputed. This issue affects every overhead and latency number in the paper.
- [Appendix A, Lemma 1 and Eq. (17)] The proof of Lemma 1 bounds the signed difference [ε_j^t]_i = [δw_j^t]_i - cδw_j^q,t/2, not its absolute value. For a low-resolution entry of magnitude x ∈ [0, λ∥δw_j^t∥∞), the actual reconstruction error is |x - cδw_j^q,t/2|, whose supremum is cδw_j^q,t/2, which can approach ∥δw_j^t∥∞/2 when the smallest high-resolution element q_t is close to ∥δw_j^t∥∞. For example, with λ=0.1, ∥δw_j^t∥∞=1, one high-resolution entry of magnitude 1 and one low-resolution entry of magnitude 0.05, q_t≈1 and the reconstruction error for the low entry is about 0.45, far above the claimed c_j≈0.05. Therefore Lemma 1 does not follow, and the Kc_max^2 term in Proposition 1, Eq. (10), is unsupported.
- [Proposition 1, Eq. (10)] The proof is deferred to Reddi et al. [22] with 'details omitted due to limited space.' This is not a direct substitution: the quantized update is not unbiased in general, and the stated conditions I–II of [22] are not verified for this quantizer. Moreover, as printed, Φ contains a term proportional to T (2LT σ̄^2), so the right-hand side of Eq. (10) tends to a nonzero constant as T grows; the claimed convergence to a stationary point does not follow from the displayed expression. A self-contained proof or a precise reduction that checks all assumptions is required.
- [Section IV, Tables II–III] The central empirical claims—matching full-precision FL accuracy and the 10% accuracy margin over benchmarks—are based on single-point accuracy numbers with no reported seeds, variance, or error bars. Table III's T_max values (e.g., 27 vs 17 for our power control vs LAQ) are the basis for the latency-budget comparison, but without repeated runs it is not possible to assess whether the differences are statistically meaningful. The absence of code further hampers reproducibility. These tables should be regenerated with multiple seeds and confidence intervals, and the claims adjusted if the side-information overhead from Major Comment 1 changes the bit counts.
minor comments (5)
- [Eq. (2)] The local update is printed as w_j^{h,t} ← −w_j^{h−1,t} − α ...; this appears to be a sign typo. The standard AdaGrad local update should be w_j^{h,t} ← w_j^{h−1,t} − α ...; as printed, the local iterate alternates in sign and is inconsistent with the rest of the paper.
- [Section II-C and Algorithm 1] The notation b_j^t is used both for the d-dimensional bit vector in Eq. (7) and for the scalar total number of bits in Section II-C and Algorithm 1; please use different symbols (e.g., β_j^t for the vector) to avoid ambiguity.
- [Abstract and Section IV] The abstract claims 'at least 93%' overhead reduction, while Section IV states 'at least r̄=96%' for Table II; these numbers should be reconciled and the parameter configurations (λ, b_j) behind each claim should be stated explicitly.
- [Algorithm 1, lines 10–11] Line 11 says users send b_j^t with b_j^p,t = ⌈log2(b_j^t)⌉ bits, but b_j^t is already defined as the scalar bit count; the purpose of this transmission and the distinction between the bit count and its encoding should be clarified.
- [Section IV and Table II] The method's overhead savings rest on the sparsity of local gradients invoked in Eq. (6)–(7), but only one CNN architecture is tested. The paper should at least discuss sensitivity to architecture and dataset, or add a robustness experiment, since the measured s values in Table II may not transfer to other models.
Circularity Check
No significant circularity: the accuracy and overhead results follow from measured sparsity and a directly solved power-control optimization, not from fitted or assumed outcomes.
full rationale
The paper's derivation chain is self-contained and non-circular. The mixed-resolution quantization scheme is defined in Eq. (6)-(8) with an explicit bit budget b_j^t = d(b_j s_j^t + 1 - s_j^t) + 32; the reported overhead reduction is an accounting identity evaluated with the measured sparsity ratio s, not a prediction fitted to the accuracy target. Lemma 1 proves a quantization-error bound directly from the quantization grid and threshold, and Proposition 1 explicitly adapts the external convergence theorem of Reddi et al. [22] by substituting the quantized gradient with a bounded error; this is a proof adaptation with restated assumptions, not a circular reuse of the conclusion. The power-control problem in Eq. (14) is a stated min-max latency objective solved by bisection and linear programming, so the computed powers are not assumed into the result. Benchmark comparisons against AQUILA, LAQ, Top-q, Dinkelbach, and max-sum rate use external methods and the paper's own bit-counting formula; no fitted parameter is renamed as a prediction. The paper's self-citations ([24] for CFmMIMO uplink details and [27] for the computation-time model) are supporting modeling references, and the uplink rate expression is attributed to the external [25, Th. 2]; they are not load-bearing for the central accuracy or overhead claim. The possible non-unique-decodability of the variable-length codewords in Eq. (7)-(8) is an implementability/correctness issue, not a circularity of the derivation, and is therefore not scored here.
Assumptions & free parameters
free parameters (4)
- Magnitude ratio threshold lambda_j =
0.05, 0.2, 0.4 in simulations
- High-resolution bit depth b_j =
4 or 10 bits
- AdaGrad step size alpha and local iterations L =
L in {3,5}, alpha not reported
- Latency model constants nu_j, a_j, and dataset size =
nu_j=20 cycles/s, a_j=1e6 cycles/sample, 5e4 samples
assumptions (5)
- standard math Convergence theorem for adaptive federated optimization in Reddi et al. [22] with Assumptions 1-3 and conditions I-II.
- domain assumption Uplink rate formula (4)-(5) from Ngo et al. [25] under independent Rayleigh fading, max-ratio combining, and imperfect channel estimation.
- domain assumption Spatial sparsity of local gradients is sufficient for sign-only representation of below-threshold entries.
- domain assumption Downlink and fronthaul transmissions are error-free; only the uplink is resource-limited.
- standard math Quantization error bound in Lemma 1 is valid as stated.
invented entities (1)
-
None
Cite this review
Pith. "Pith review of Adaptive Quantization Resolution and Power Control for Federated Learning over Cell-free Networks." pith.science (2026). https://pith.science/paper/34EESAW5
@misc{pith2026241210878,
author = {Pith},
title = {Pith review of: Adaptive Quantization Resolution and Power Control for Federated Learning over Cell-free Networks},
year = {2026},
howpublished = {\url{https://pith.science/paper/34EESAW5}},
note = {Machine review of arXiv:2412.10878}
}
read the original abstract
Federated learning (FL) is a distributed learning framework where users train a global model by exchanging local model updates with a server instead of raw datasets, preserving data privacy and reducing communication overhead. However, the latency grows with the number of users and the model size, impeding the successful FL over traditional wireless networks with orthogonal access. Cell-free massive multiple-input multipleoutput (CFmMIMO) is a promising solution to serve numerous users on the same time/frequency resource with similar rates. This architecture greatly reduces uplink latency through spatial multiplexing but does not take application characteristics into account. In this paper, we co-optimize the physical layer with the FL application to mitigate the straggler effect. We introduce a novel adaptive mixed-resolution quantization scheme of the local gradient vector updates, where only the most essential entries are given high resolution. Thereafter, we propose a dynamic uplink power control scheme to manage the varying user rates and mitigate the straggler effect. The numerical results demonstrate that the proposed method achieves test accuracy comparable to classic FL while reducing communication overhead by at least 93% on the CIFAR-10, CIFAR-100, and Fashion-MNIST datasets. We compare our methods against AQUILA, Top-q, and LAQ, using the max-sum rate and Dinkelbach power control schemes. Our approach reduces the communication overhead by 75% and achieves 10% higher test accuracy than these benchmarks within a constrained total latency budget.
Figures
Forward citations
Cited by 1 Pith paper
-
Accelerating Energy-Efficient Federated Learning in Cell-Free Networks with Adaptive Quantization
An adaptive element-wise quantization scheme and power allocation method for federated learning over cell-free massive MIMO improves test accuracy by up to 7-19% under equal energy and latency budgets.
Reference graph
Works this paper leans on
-
[22]
Adaptive federated optimization,
S. Reddi et al. , “Adaptive federated optimization,” arXiv preprint arXiv:2003.00295, 2020
arXiv 2003
-
[1]
Federated Learning: Strategies for improving com- munication efficiency,
J. Kone ˇcn`y et al. , “Federated Learning: Strategies for improving com- munication efficiency,” arXiv preprint arXiv:1610.05492 , 2016
arXiv 2016
-
[2]
Foundations of user-centric cell-free Massive MIMO,
O. T. Demir and E. Bj ¨ornson, “Foundations of user-centric cell-free Massive MIMO,” Foundations and Trends in Signal Processing, vol. 14, no. 3-4, pp. 162–472, 2021
work page 2021
-
[3]
Wireless communications for collaborative Federated Learning,
M. Chen et al. , “Wireless communications for collaborative Federated Learning,” IEEE Communications Magazine , vol. 58, pp. 48–54, 2020
2020
-
[4]
Communication-efficient Federated Learning with adap- tive quantization,
Y . Mao et al., “Communication-efficient Federated Learning with adap- tive quantization,” ACM Trans. Intell. Syst. Technol. , vol. 13, no. 4, pp. 1–26, 2022
work page 2022
-
[5]
Adaptive quantization of model updates for communication-efficient Federated Learning,
D. Jhunjhunwala et al. , “Adaptive quantization of model updates for communication-efficient Federated Learning,” in 2021 IEEE ICASSP , pp. 3110–3114
2021
-
[6]
Joint compression and deadline optimization for wireless Federated Learning,
M. Zhang et al. , “Joint compression and deadline optimization for wireless Federated Learning,” IEEE Trans. Mobile Comput. , vol. 23, no. 7, pp. 7939–7951, 2024
work page 2024
-
[7]
FedDQ: Communication-efficient Federated Learning with descending quantization,
L. Qu et al., “FedDQ: Communication-efficient Federated Learning with descending quantization,” in 2022 IEEE GLOBECOM , pp. 281–286
2022
Show all 27 references
-
[8]
AQUILA: Communication efficient Federated Learning with adaptive quantization in device selection strategy,
Z. Zhao et al., “AQUILA: Communication efficient Federated Learning with adaptive quantization in device selection strategy,” IEEE Trans. Mobile Comput., vol. 23, no. 6, pp. 7363–7376, 2024
2024
-
[9]
Lazily Aggregated Quantized Gradient (LAQ) innovation for communication-efficient Federated Learning,
J. Sun et al., “Lazily Aggregated Quantized Gradient (LAQ) innovation for communication-efficient Federated Learning,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 44, no. 4, pp. 2031–2044, 2022
2022
-
[10]
A-LAQ: Adaptive Lazily Aggregated Quantized Gradient,
A. Mahmoudi et al. , “A-LAQ: Adaptive Lazily Aggregated Quantized Gradient,” in 2022 IEEE Globecom Workshops , pp. 1828–1833
2022
-
[11]
DAdaQuant: Doubly-adaptive quantization for communication-efficient Federated Learning,
R. H ¨onig et al. , “DAdaQuant: Doubly-adaptive quantization for communication-efficient Federated Learning,” in ICML, 2022, pp. 8852– 8866
2022
-
[12]
Joint resource allocation to minimize execution time of Federated Learning in cell-free Massive MIMO,
T. T. Vu et al., “Joint resource allocation to minimize execution time of Federated Learning in cell-free Massive MIMO,” IEEE Internet Things J., vol. 9, no. 21, pp. 21 736–21 750, 2022
2022
-
[13]
Cell-free Massive MIMO for wireless Federated Learning,
——, “Cell-free Massive MIMO for wireless Federated Learning,” IEEE Trans. Wireless Commun., vol. 19, no. 10, pp. 6377–6392, 2020
2020
-
[14]
Straggler effect mitigation for Federated Learning in cell-free Massive MIMO,
——, “Straggler effect mitigation for Federated Learning in cell-free Massive MIMO,” in 2021 IEEE ICC , pp. 1–6
2021
-
[15]
How does cell-free Massive MIMO support multiple Federated Learning groups?
——, “How does cell-free Massive MIMO support multiple Federated Learning groups?” in 2021 IEEE SPAWC, pp. 401–405
2021
-
[16]
FedAQ: Communication-efficient Federated Edge Learning via joint uplink and downlink adaptive quantization,
L. Qu et al., “FedAQ: Communication-efficient Federated Edge Learning via joint uplink and downlink adaptive quantization,” arXiv preprint arXiv:2406.18156, 2024
2024 arXiv
-
[17]
SAFARI: Sparsity-enabled Federated Learning with limited and unreliable communications,
Y . Mao et al. , “SAFARI: Sparsity-enabled Federated Learning with limited and unreliable communications,” IEEE Trans. Mobile Comput. , vol. 23, no. 5, pp. 4819–4831, 2024
2024
-
[18]
Machine learning at the wireless edge: Distributed stochastic gradient descent Over-the-Air,
M. M. Amiri et al., “Machine learning at the wireless edge: Distributed stochastic gradient descent Over-the-Air,” in 2019 IEEE ISIT, pp. 1432– 1436
2019
-
[19]
Vector quantized compressed sensing for communication- efficient Federated Learning,
Y . Oh et al., “Vector quantized compressed sensing for communication- efficient Federated Learning,” in 2022 IEEE Globecom Workshops , pp. 365–370
2022
-
[20]
Gradient sparsification for communication-efficient distributed optimization,
J. Wangni et al. , “Gradient sparsification for communication-efficient distributed optimization,” in Advances in Neural Information Processing Systems, 2018, pp. 1299–1309
2018
-
[21]
Energy-efficient power control: A look at 5G wireless technologies,
A. Zappone et al., “Energy-efficient power control: A look at 5G wireless technologies,” IEEE Trans. Signal Process. , vol. 64, pp. 1668–1683, 2015
2015
-
[23]
Adadelta: An adaptive learning rate method,
M. D. Zeiler, “Adadelta: An adaptive learning rate method,” arXiv preprint arXiv:1212.5701, 2012
2012 arXiv
-
[24]
Joint energy and latency optimization in Federated Learning over cell-free massive MIMO networks,
A. Mahmoudi et al., “Joint energy and latency optimization in Federated Learning over cell-free massive MIMO networks,” in2024 IEEE WCNC
-
[25]
Cell-free Massive MIMO versus small cells,
H. Q. Ngo et al., “Cell-free Massive MIMO versus small cells,” IEEE Trans. Wireless Commun., vol. 16, pp. 1834–1850, 2017
2017
-
[26]
Communication-efficient learning of deep networks from decentralized data,
B. McMahan et al., “Communication-efficient learning of deep networks from decentralized data,” in Artificial intelligence and statistics. PMLR, 2017, pp. 1273–1282
2017
-
[27]
FedCau: A proactive stop policy for commu- nication and computation efficient Federated Learning,
A. Mahmoudi et al. , “FedCau: A proactive stop policy for commu- nication and computation efficient Federated Learning,” IEEE Trans. Wireless Commun., pp. 1–1, 2024
2024
Reviewed August 11, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.