Empirical Analysis of Asynchronous Federated Learning on Heterogeneous Devices: Efficiency, Fairness, and Privacy Trade-offs

Ali Balador; Francesco Flammini; Iraklis Symeonidis; Samaneh Mohammadi

REVIEW 3 major objections 4 minor 27 references

Empirical Analysis of Asynchronous Federated Learning on Heterogeneous Devices: Efficiency, Fairness, and Privacy Trade-offs

T0 review · 3 major / 4 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 Asynchronous federated learning converges up to 10x faster, but the privacy burden falls unevenly on high-end devices, which incur up to 5x higher privacy loss.

desk verdict Empirically interesting and credible on efficiency/fairness, but the privacy-disparity numbers don't cohere with the paper's own accounting, so treat the 5x claim with caution. read the letter →

arxiv 2505.07041 v1 pith:7VODBULU submitted 2025-05-11 cs.DC cs.AIcs.LG

Samaneh Mohammadi , Iraklis Symeonidis , Ali Balador , Francesco Flammini This is my paper · ORCID

classification cs.DCcs.AIcs.LG

keywords federatedlearningasynchronousaggregationdeviceheterogeneitylocaldifferentialprivacymomentsaccountantfairnessspeechemotionrecognition

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

This paper argues that the efficiency gain of asynchronous federated learning comes with a hidden, systematic fairness and privacy cost on heterogeneous hardware. Using a physical testbed of five devices running a speech-emotion-recognition task, it compares synchronous FedAvg with staleness-aware FedAsync under local differential privacy, tracking per-client cumulative privacy loss with the Moments Accountant. It finds that FedAsync reaches target accuracy about 9-10x faster in wall-clock time, but high-end clients contribute 6-10x more updates and accumulate up to roughly 5x higher privacy loss, while low-end clients lose more local accuracy under the same noise. The authors conclude that static uniform privacy mechanisms are inadequate and that aggregation, staleness handling, and privacy budgets should be tuned jointly based on client capacity and participation.

What carries the argument

The load-bearing mechanism is the coupling of two standard components: staleness-aware asynchronous aggregation (FedAsync's decay factor $\alpha_k = \alpha/(1+\tau_k)$, which down-weights old updates) and per-client Moments Accountant privacy accounting ($\varepsilon$ computed from a fixed noise scale, sampling probability $q$, and the number of rounds the client actually contributes). The argument runs through this coupling: because faster devices send more updates with low staleness, they receive larger aggregation weights and accumulate proportionally larger $\varepsilon$; slower devices send fewer, staler updates, so they get less influence and smaller $\varepsilon$, yet suffer more noise-induced accuracy loss. The trade-off is summarized by varying $\alpha$, which trades convergence speed against participation skew and privacy disparity.

What would settle it

Recomputing each Table 3 epsilon from the paper's measured participation percentages using Eqs. (7)-(8) with $\sigma$ fixed would settle it: if a device's epsilon is not proportional to how often it actually updated, or if FedAvg's epsilon is not identical across devices, the claim that privacy loss tracks participation fails.

Watch

Extended reading notes

Core claim

The central claim, on the paper's own terms, is that under realistic device heterogeneity asynchronous aggregation converts the straggler problem into an inequality problem: the same staleness-aware weighting that makes FedAsync fast also makes it unfair and privacy-uneven. Empirically, at alpha=0.6 high-end devices account for about 80% of updates while low-end devices fall below 5%, and the cumulative privacy loss at sigma=0.5 reaches roughly 35 for the fastest device versus below 10 for the slowest. Because each device's update count drives its Moments Accountant budget, the privacy burden tracks participation exactly, and uniform LDP noise does not equalize outcomes: low-end devices, whose few updates are stale and down-weighted, show up to about 16% local accuracy degradation while high-end devices degrade by roughly 6%.

Load-bearing premise

The headline privacy-disparity numbers rest on the assumption that each device's reported epsilon was driven by that device's real update count, an assumption the paper's own FedAvg row (identical epsilon for all five devices despite documented low-end dropouts) leaves unverified.

Editorial extensions

If this is right

If the paper is right, reported convergence speedups of asynchronous FL should be read together with the per-device participation skew; the 10x wall-clock gain is purchased by letting fast hardware dominate the update stream.
Uniform local differential privacy noise does not deliver uniform privacy or uniform utility; low-end clients bear disproportionate accuracy loss under the same noise scale.
The aggregation strength $\alpha$ acts as a fairness-privacy dial: lowering $\alpha$ narrows participation and privacy gaps at the cost of convergence speed.
Per-client privacy accounting that uses real update counts would show the same disparity in any asynchronous FL deployment where participation is hardware-bound.
Systems that need both speed and equity should weight updates by capacity or staleness while calibrating noise per client, rather than applying one-size-fits-all LDP.

Reading between the lines

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

The paper does not state this, but its epsilon equations imply that the privacy disparity is a participation fact rather than a hardware fact: any asynchronous mechanism that changes who sends updates changes who pays the privacy cost.
A direct extension of the authors' own suggestion would be to inverse-calibrate per-client noise, adding more noise to frequent high-end contributors and less to rare low-end ones, and test whether the epsilon spread narrows without pushing global accuracy below the FedAvg level.
Because the paper reports participation, staleness, and epsilon together, one could define a composite privacy-fairness index equal to epsilon divided by influence; under the reported numbers this index would be near-constant if the accountant tracks participation, which would make the disparity purely a weighting artifact.
Comparing with client-level differential privacy instead of LDP would isolate whether low-end accuracy loss comes from per-message noise or from stale-model training; the paper does not run this control.

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.

Desk Editor's Note

read the letter

Quick take: this is a useful empirical paper, not a breakthrough. The physical testbed (five hardware tiers, SER, real FedAsync/FedAvg) is a nice change from simulation, and the qualitative message—async FL accelerates convergence but skews participation and privacy burden toward fast hardware—is almost certainly right and worth saying.

What's new: per-client privacy accounting under FedAsync using the Moments Accountant on real devices, with participation percentages, staleness, and per-device accuracy degradation. The convergence speedup (about 9-10x to 75% accuracy) and the participation skew (high-end devices contributing 30-42% vs 4-8% for low-end) are concrete measurements that support the fairness concern. Those parts are solid.

The soft spot is the privacy accounting. In Section 3.2, epsilon is composed over a device's number of local updates; with fixed sigma and q, per-client epsilon should roughly scale with update count. But Table 3's epsilon ratios don't match the participation percentages in Fig 5. At alpha=0.2, T5/T1 participation is 31.87/8.24 ≈ 3.9, while the epsilon ratio is 31.75/12.57 ≈ 2.5. At alpha=0.6, participation ratio is about 9.4 but epsilon ratio is 5.2. That gap is too large to excuse as accountant rounding. More telling, the FedAvg row reports a single epsilon for all devices even though Section 4.2.1 documents low-end dropouts; per-client accounting should lower their composed epsilon. That suggests the accountant wasn't actually driven by measured update counts. Without code or per-client logs, the headline '5x privacy disparity' is not supported by the data as presented.

Other, smaller issues: the accuracy degradation baseline in Table 3 isn't defined, and no code or data are released, which is a miss for an empirical paper.

Who this is for: people working on heterogeneous FL, fairness, or DP in FL will find the experimental setup useful and the trade-off story worth citing as motivation, but only after the authors either fix the accounting or release artifacts showing it matches participation. The paper deserves peer review, but I'd want the accounting discrepancy addressed.

Referee Report

3 major / 4 minor

Summary. This paper reports an empirical comparison of synchronous FedAvg and staleness-aware FedAsync on a physical testbed of five edge devices (Raspberry Pi 3/4, NXP HummingBoard) for speech emotion recognition on CREMA-D. Local updates are perturbed with Gaussian noise under a DP-SGD-style mechanism, and per-client privacy loss is tracked with a custom Moments Accountant. The paper claims that FedAsync reaches 75% accuracy roughly 9-10x faster in wall-clock time than FedAvg, that high-end devices contribute 6-10x more updates and incur up to 5x higher per-client epsilon, and that low-end devices suffer greater accuracy degradation. The efficiency and participation measurements are plausible and largely internally consistent, but the privacy-loss accounting is not reconciled with the measured participation and dropout behavior, so the quantitative privacy-disparity claims are not supported as reported.

Significance. If the privacy accounting is corrected, the central message—that asynchronous FL improves wall-clock convergence while shifting both influence and privacy burden toward high-end devices—would be a useful empirical contribution for systems designers choosing aggregation and privacy mechanisms. The paper's strengths are the physical testbed spanning five hardware tiers, multi-seed averaging, direct wall-clock convergence measurements, and explicit per-device participation, staleness, and accuracy breakdowns. The fairness and efficiency findings are more solid than the privacy findings. However, the paper does not release code, per-client update-count logs, or the custom accountant implementation, and the privacy-loss numbers are the part of the paper that most needs independent verification.

major comments (3)

[Section 4.2.1, Table 2; Section 4.2.3, Table 3; Section 3.2, Eqs. (7)-(8)] The FedAvg row in Table 3 reports a single epsilon for 'All Devices' at each sigma, yet Section 4.2.1 and Table 2 document that HW T1 dropped out three times and HW T2 dropped out twice during FedAvg training. Under the per-client accountant of Eqs. (7)-(8), a client that misses rounds must have strictly fewer composed moments and hence strictly lower cumulative epsilon than a client that never drops out. Equal epsilon for all devices is therefore only consistent with an accountant fed a nominal fixed round count rather than measured per-client participation. Since the same accounting pipeline is used for FedAsync, the per-client epsilon values in Table 3 are not shown to follow from the update counts measured in Section 4.2.2. This directly undermines the headline claim of up to 5x higher privacy loss for high-end devices; the authors should release per-client update-count logs and the accountant code, and recompute Table 3 from the measured participation, or explicitly reconcile the FedAvg row with the documented dropouts.
[Section 3.2, Eqs. (4)-(8); Algorithm 1, lines 8-16] The privacy accounting adds one moment mu_t per client round, but DP-SGD as written in Eqs. (4)-(6) applies clipping and Gaussian noise to every mini-batch. With E=1, B=128, and about 941 training samples per client, each round contains roughly eight noised mini-batch steps, and the Moments Accountant for DP-SGD must compose over those mini-batch steps rather than over whole rounds. If mu_t in Eq. (7) is intended to summarize an entire round, the paper must explain how the composition over mini-batches is folded into a single mechanism; as written, the reported epsilon values appear to undercount privacy loss by approximately the number of mini-batches per round. All absolute epsilon values in Table 3 are therefore questionable independently of the participation-mismatch issue.
[Section 4.2.2, Fig. 5; Section 4.2.3, Table 3] The epsilon ratios in Table 3 do not track the participation ratios in Fig. 5 under the stated accountant. At alpha=0.2 and sigma=0.5, HW T5 has 31.87% participation versus 8.24% for HW T1 (ratio 3.87), while the reported epsilon ratio is 31.75/12.57 = 2.53; at alpha=0.6, the participation ratio is 42.34/4.50 = 9.41 but the epsilon ratio is 35.12/6.79 = 5.17. Sublinear RDP composition might explain the divergence, but the paper does not provide the per-update RDP parameters or the accountant trace needed to verify this. Because the headline privacy ratio is derived from these numbers, the authors should show the step-by-step accountant output for at least one alpha and one sigma value.

minor comments (4)

[Section 6 and Abstract] The conclusion states that high-end clients experience 'up to 6x greater privacy loss', while the abstract and Section 4.2.4 state 'up to 5x'; Table 3 gives a maximum ratio of about 5.17, so the multiplier should be harmonized across the paper.
[Section 4.1.4] The paper says FedAsync runs 'continuous updates' until convergence, but the total number of asynchronous updates and the per-client update counts are never stated; Table 3's epsilon values cannot be reproduced without these counts. A table or appendix listing update counts for each device and each alpha would resolve this.
[Figure 4] The inset in Figure 4 uses sparse tick labels such as 67, 467, 867, 1267 without a clear indication of the scale; the inset axes should be labeled explicitly and the tick format made readable.
[Section 3.2, Eq. (7)] Equation (7) does not precisely define D and D' for the local mechanism; the authors should state explicitly that they are neighboring local datasets differing in one sample, and clarify whether privacy is accounted at the sample level or the client level.

Circularity Check

1 steps flagged · score 6.0 of 10

Privacy-loss disparity is a deterministic rescaling of the measured participation imbalance, so the headline 5x privacy-loss claim reduces by construction to the update-count result; the FedAvg control row also conflicts with documented dropouts.

self definitional [Section 3.2, Eqs. (7)-(8); Algorithm 1 lines 14-17; Table 3; Section 4.2.1]
"The cumulative moment and resulting privacy budget are: µ(λ)=Σ_{t=1}^T µ_t(λ), ε=min_λ((µ(λ)−log δ)/λ). Algorithm 1: Update µ^{(k)}(λ)←µ^{(k)}(λ)+µ^{(k)}_t(λ). Table 3: FedAvg All Devices 26.55±0.47. Section 4.2.1: HW T1 ... three observed dropouts; HW T2 ... two dropouts."

With σ and q fixed (Section 4.1.4: σ∈{0.5,1,1.5,2}, q≈0.136), every per-round log moment µ_t is the same function for each device; the only client-specific variable in Eq. (8) is the number of rounds T_k over which a client's moments are accumulated. Table 3's per-client ε under FedAsync is therefore a deterministic, monotone transform of the participation percentages already measured in Fig. 5. The abstract's 'up to 5x higher privacy loss' is thus the '6–10x more updates' result restated through the accountant's composition formula, not an independent empirical finding. The FedAvg control row reinforces that the accountant was not actually driven by measured participation: Section 4.2.1 reports HW T1/T2 dropouts, yet Table 3 lists one identical ε for 'All Devices'.

full rationale

The paper's efficiency and fairness findings (10x faster convergence, participation skew at different α) are empirical, self-contained, and not circular. The per-client privacy-loss disparity, however, is computed by Eq. (8) as a sum of per-step moments over each client's updates. Since the noise scale and sampling probability are held constant across clients, the reported ε values are a monotone function of the very participation counts measured in Fig. 5; the '5x higher privacy loss' claim therefore carries no empirical content beyond the already-measured update-count imbalance. The FedAvg row heightens this concern: Table 3 reports a single ε for all devices even though Section 4.2.1 documents dropouts for HW T1 and HW T2, indicating the accountant was not transparently driven by per-client measured participation. No load-bearing self-citation or imported uniqueness theorem appears; references [3] and [16] are background only. Because the central privacy-disparity claim reduces by construction to the participation input while the efficiency/fairness claims are independent, the circularity is partial, not total.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

Free parameters are experimental hyperparameters, not fitted constants. No new entities are postulated. The central caveat is that the per-client epsilon values are deterministic outputs of the DP accountant given update counts, so the privacy asymmetry reported in Table 3 is built into the accounting method rather than discovered independently of participation.

free parameters (2)

Staleness decay factor alpha = 0.2, 0.4, 0.6
Server-side aggregation weight in Eq. 10, chosen by hand. The fairness and convergence results in Figs 4-5 and Table 3 are parameterized by it; the headline speedup is reported at alpha=0.4.
LDP Gaussian noise scale sigma = 0.5, 1.0, 1.5, 2.0
Noise scale in Eq. 5, chosen by hand. All per-client privacy-loss values and utility-loss comparisons in Table 3 depend on it.

assumptions (3)

domain assumption Per-client privacy loss is computed by composing one DP-SGD step per local epoch with sampling probability q=B/|D_k| and the same sigma for every device, and this composition is valid in the asynchronous setting.
Section 3.2 and Algorithm 1 lines 14-18; all epsilon values in Table 3 rest on this. The moments accountant's validity under stale, asynchronous, non-Poisson update patterns is assumed, not proved.
domain assumption The fixed batch size B=128 with q=B/|D_k| approximately 0.136 accurately models the privacy sampling in the local data loaders.
Section 4.1.4. DP-SGD accountants assume Poisson subsampling; the paper does not describe whether the implementation uses Poisson sampling or a fixed-order loader, so this is an approximation that can affect epsilon bounds.
domain assumption The five CREMA-D partitions are IID and equal-sized, so performance divergence between devices is attributed to hardware and aggregation rather than data distribution.
Section 4.1.3 states an IID split but does not quantify distributional similarity; the fairness conclusions depend on this assumption.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Empirical Analysis of Asynchronous Federated Learning on Heterogeneous Devices: Efficiency, Fairness, and Privacy Trade-offs." pith.science (2026). https://pith.science/paper/7VODBULU

@misc{pith2026250507041,
  author       = {Pith},
  title        = {Pith review of: Empirical Analysis of Asynchronous Federated Learning on Heterogeneous Devices: Efficiency, Fairness, and Privacy Trade-offs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7VODBULU}},
  note         = {Machine review of arXiv:2505.07041}
}

read the original abstract

Device heterogeneity poses major challenges in Federated Learning (FL), where resource-constrained clients slow down synchronous schemes that wait for all updates before aggregation. Asynchronous FL addresses this by incorporating updates as they arrive, substantially improving efficiency. While its efficiency gains are well recognized, its privacy costs remain largely unexplored, particularly for high-end devices that contribute updates more frequently, increasing their cumulative privacy exposure. This paper presents the first comprehensive analysis of the efficiency-fairness-privacy trade-off in synchronous vs. asynchronous FL under realistic device heterogeneity. We empirically compare FedAvg and staleness-aware FedAsync using a physical testbed of five edge devices spanning diverse hardware tiers, integrating Local Differential Privacy (LDP) and the Moments Accountant to quantify per-client privacy loss. Using Speech Emotion Recognition (SER) as a privacy-critical benchmark, we show that FedAsync achieves up to 10x faster convergence but exacerbates fairness and privacy disparities: high-end devices contribute 6-10x more updates and incur up to 5x higher privacy loss, while low-end devices suffer amplified accuracy degradation due to infrequent, stale, and noise-perturbed updates. These findings motivate the need for adaptive FL protocols that jointly optimize aggregation and privacy mechanisms based on client capacity and participation dynamics, moving beyond static, one-size-fits-all solutions.

Figures

Figures reproduced from arXiv: 2505.07041 by the authors.

**Figure 1.** Federated learning under device heterogeneity. Comparison of synchronous and asynchronous FL for speech emotion recognition. Local update privacy is ensured by DP-SGD, with per-client privacy loss tracked by the moments accountant. Devices range from low-end (HW T1) to high-end (HW T5), reflecting realistic hardware variability. 2.4 Speech Emotion Recognition SER classifies emotional states from speech signals by an… view at source ↗

**Figure 2.** FL testbed with clients comprising heterogeneous devices (HW T1–T5) To accurately reflect real-world SER deployment scenarios (as outlined in Section 2.4), we built an FL testbed comprising five heterogeneous edge devices and a central server. The devices span five distinct hardware configurations (HW T1–T5), varying in computational capacity (CPU, RAM), as detailed in [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** FL training performance per round across heterogeneous devices. HW T3 demonstrated intermediate performance, with training time and exchange latency approximately 3–4× faster than low-end devices and 3–4× slower than high-end devices. High-end devices (HW T4, HW T5) achieved the best results, showing smooth accuracy improvements with minimal test accuracy fluctuations, training durations of 65–75 s, and exchange lat… view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Convergence time of FedAvg vs. FedAsync with and without staleness-aware aggregation under device heterogeneity. Notably, we also observe the role of stalenessaware aggregation. While both FedAsync variants achieve substantial speedups, staleness-aware FedAsync deliv…

**Figure 5.** Figure 5: Global and per-device accuracy trajectories, along with participation percentages (PP), in asynchronous FL with staleness-aware weighting under varying aggregation strengths α. Higher α accelerates convergence but exacerbates representational skew by diminishing parti…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 20 canonical work pages

[1]

Communication- efficient learning of deep networks from decentralized data

Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas. Communication- efficient learning of deep networks from decentralized data. InArtificial intelligence and statistics, pages 1273–
[2]

Advances and open problems in federated learning.Foundations and Trends® in Machine Learning, 14(1–2):1–210, 2021

Peter Kairouz, H Brendan McMahan, Brendan Avent, Aur ´elien Bellet, Mehdi Bennis, Arjun Nitin Bhagoji, Kallista Bonawitz, Zachary Charles, Graham Cormode, Rachel Cummings, et al. Advances and open problems in federated learning.Foundations and Trends® in Machine Learning, 14(1–2):1–210, 2021

work page 2021
[3]

Balancing privacy and performance in federated learning: a systematic literature review on methods and metrics.Journal of Parallel and Distributed Computing, page 104918, 2024

Samaneh Mohammadi, Ali Balador, Sima Sinaei, and Francesco Flammini. Balancing privacy and performance in federated learning: a systematic literature review on methods and metrics.Journal of Parallel and Distributed Computing, page 104918, 2024

work page 2024
[4]

Client selection for federated learning with heterogeneous resources in mobile edge

Takayuki Nishio and Ryo Yonetani. Client selection for federated learning with heterogeneous resources in mobile edge. InICC 2019-2019 IEEE international conference on communications (ICC), pages 1–7. IEEE, 2019

work page 2019
[5]

Federated learning with buffered asynchronous aggregation

John Nguyen, Kshitiz Malik, Hongyuan Zhan, Ashkan Yousefpour, Mike Rabbat, Mani Malek, and Dzmitry Huba. Federated learning with buffered asynchronous aggregation. InInternational Conference on Artificial Intelligence and Statistics, pages 3581–3607. PMLR, 2022

work page 2022
[6]

Tifl: A tier-based federated learning system

Zheng Chai, Ahsan Ali, Syed Zawad, Stacey Truex, Ali Anwar, Nathalie Baracaldo, Yi Zhou, Heiko Ludwig, Feng Yan, and Yue Cheng. Tifl: A tier-based federated learning system. InProceedings of the 29th international symposium on high-performance parallel and distributed computing, pages 125–136, 2020. 14

work page 2020
[7]

Asynchronous federated optimization.arXiv preprint arXiv:1903.03934, 2019

Cong Xie, Sanmi Koyejo, and Indranil Gupta. Asynchronous federated optimization.arXiv preprint arXiv:1903.03934, 2019

arXiv 1903
[8]

Local differential privacy for deep learning.IEEE Internet of Things Journal, 7(7):5827– 5842, 2019

Pathum Chamikara Mahawaga Arachchige, Peter Bertok, Ibrahim Khalil, Dongxi Liu, Seyit Camtepe, and Mo- hammed Atiquzzaman. Local differential privacy for deep learning.IEEE Internet of Things Journal, 7(7):5827– 5842, 2019

work page 2019

Show all 27 references

[9]

Deep learning with differential privacy

Martin Abadi, Andy Chu, Ian Goodfellow, H Brendan McMahan, Ilya Mironov, Kunal Talwar, and Li Zhang. Deep learning with differential privacy. InProceedings of the 2016 ACM SIGSAC conference on computer and communications security, pages 308–318, 2016

2016
[10]

Linear queries estimation with local differential privacy

Raef Bassily. Linear queries estimation with local differential privacy. InThe 22nd International Conference on Artificial Intelligence and Statistics, pages 721–729. PMLR, 2019

2019
[11]

A review on speech emotion recognition: a survey, recent advances, challenges, and the influence of noise.Neurocomputing, 568:127015, 2024

Swapna Mol George and P Muhamed Ilyas. A review on speech emotion recognition: a survey, recent advances, challenges, and the influence of noise.Neurocomputing, 568:127015, 2024

2024
[12]

Enhancing smart home design with ai models: A case study of living spaces implementation review.Energies, 16(6):2636, 2023

Amjad Almusaed, Ibrahim Yitmen, and Asaad Almssad. Enhancing smart home design with ai models: A case study of living spaces implementation review.Energies, 16(6):2636, 2023

2023
[13]

Automatic speech emotion recognition using machine learning.Social Media and Machine Learning [Working Title], 2019

Leila Kerkeni, Youssef Serrestou, Mohamed Mbarki, Kosai Raoof, Mohamed Ali Mahjoub, and Catherine Cleder. Automatic speech emotion recognition using machine learning.Social Media and Machine Learning [Working Title], 2019

2019
[14]

Swadha Gupta, Parteek Kumar, and Raj Kumar Tekchandani. Facial emotion recognition based real-time learner engagement detection system in online learning context using deep learning models.Multimedia Tools and Applications, 82(8):11365–11394, 2023

2023
[15]

Privacy implications of voice and speech analysis–information disclosure by inference.Privacy and Identity Management

Jacob Leon Kr ¨oger, Otto Hans-Martin Lutz, and Philip Raschke. Privacy implications of voice and speech analysis–information disclosure by inference.Privacy and Identity Management. Data for Better Living: AI and Privacy: 14th IFIP WG 9.2, 9.6/11.7, 11.6/SIG 9.2. 2 Internatio...

2019
[16]

Balancing privacy and accuracy in federated learning for speech emotion recogni- tion

Samaneh Mohammadi, Mohammadreza Mohammadi, Sima Sinaei, Ali Balador, Ehsan Nowroozi, Francesco Flammini, and Mauro Conti. Balancing privacy and accuracy in federated learning for speech emotion recogni- tion. In2023 18th Conference on Computer Science and Intelligence Systems ...

2023
[17]

Differentially private federated learning: A client level perspec- tive

Robin C Geyer, Tassilo Klein, and Moin Nabi. Differentially private federated learning: A client level perspec- tive. InNeurIPS Workshop on Private Multi-Party Machine Learning, 2017

2017
[18]

Speech emotion recognition with deep convolutional neural networks.Biomedical Signal Processing and Control, 59:101894, 2020

Dias Issa, M Fatih Demirci, and Adnan Yazici. Speech emotion recognition with deep convolutional neural networks.Biomedical Signal Processing and Control, 59:101894, 2020

2020
[19]

Light-sernet: A lightweight fully convolutional neural network for speech emotion recognition.arXiv preprint arXiv:2110.03435, 2021

Arya Aftab, Shivangi Sharma, and Swati Srivastava. Light-sernet: A lightweight fully convolutional neural network for speech emotion recognition.arXiv preprint arXiv:2110.03435, 2021

2021 arXiv
[20]

Enhancing emotion recognition through federated learning: A multimodal approach with convolutional neural networks.Applied Sciences, 14(4), 2024

Nikola Simic, Haseeb Yousaf, and Emiliano Casalicchio. Enhancing emotion recognition through federated learning: A multimodal approach with convolutional neural networks.Applied Sciences, 14(4), 2024. 15

2024
[21]

Personalized federated learning with differential privacy.IEEE Internet of Things Journal, 7(10):9530–9539, 2020

Rui Hu, Yuanxiong Guo, Hongning Li, Qingqi Pei, and Yanmin Gong. Personalized federated learning with differential privacy.IEEE Internet of Things Journal, 7(10):9530–9539, 2020

2020
[22]

Flower: A friendly federated learning research framework.arXiv preprint arXiv:2007.14390, 2020

Daniel J Beutel, Taner Topal, Akhil Mathur, Xinchi Qiu, Javier Fernandez-Marques, Yan Gao, Lorenzo Sani, Kwing Hei Li, Titouan Parcollet, Pedro Porto Buarque de Gusm ˜ao, et al. Flower: A friendly federated learning research framework.arXiv preprint arXiv:2007.14390, 2020

2007 arXiv
[23]

Opacus: User-friendly differential privacy library in pytorch.arXiv preprint arXiv:2109.12298, 2021

Ashkan Yousefpour, Igor Shilov, Alexandre Sablayrolles, Davide Testuggine, Karthik Prasad, Mani Malek, John Nguyen, Sayan Ghosh, Akash Bharadwaj, Jessica Zhao, et al. Opacus: User-friendly differential privacy library in pytorch.arXiv preprint arXiv:2109.12298, 2021

2021 arXiv
[24]

Crema-d: Crowd-sourced emotional multimodal actors dataset.IEEE transactions on affective computing, 5(4):377–390, 2014

Houwei Cao, David G Cooper, Michael K Keutmann, Ruben C Gur, Ani Nenkova, and Ragini Verma. Crema-d: Crowd-sourced emotional multimodal actors dataset.IEEE transactions on affective computing, 5(4):377–390, 2014

2014
[25]

Projected federated averaging with heterogeneous differential privacy.Proceedings of the VLDB Endowment, 15(4):828–840, 2021

Junxu Liu, Jian Lou, Li Xiong, Jinfei Liu, and Xiaofeng Meng. Projected federated averaging with heterogeneous differential privacy.Proceedings of the VLDB Endowment, 15(4):828–840, 2021

2021
[26]

Online client selection for asynchronous federated learning with fairness consideration.IEEE Transactions on Wireless Communications, 22(4):2493–2506, 2022

Hongbin Zhu, Yong Zhou, Hua Qian, Yuanming Shi, Xu Chen, and Yang Yang. Online client selection for asynchronous federated learning with fairness consideration.IEEE Transactions on Wireless Communications, 22(4):2493–2506, 2022

2022
[27]

Adaptive personalized federated learning

Yuyang Deng, Mohammad Mahdi Kamani, and Mehrdad Mahdavi. Adaptive personalized federated learning. arXiv preprint arXiv:2003.13461, 2020. 16

2003 arXiv

Pith tools

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

[1] [1]

Communication- efficient learning of deep networks from decentralized data

Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas. Communication- efficient learning of deep networks from decentralized data. InArtificial intelligence and statistics, pages 1273–

[2] [2]

Advances and open problems in federated learning.Foundations and Trends® in Machine Learning, 14(1–2):1–210, 2021

Peter Kairouz, H Brendan McMahan, Brendan Avent, Aur ´elien Bellet, Mehdi Bennis, Arjun Nitin Bhagoji, Kallista Bonawitz, Zachary Charles, Graham Cormode, Rachel Cummings, et al. Advances and open problems in federated learning.Foundations and Trends® in Machine Learning, 14(1–2):1–210, 2021

work page 2021

[3] [3]

Balancing privacy and performance in federated learning: a systematic literature review on methods and metrics.Journal of Parallel and Distributed Computing, page 104918, 2024

Samaneh Mohammadi, Ali Balador, Sima Sinaei, and Francesco Flammini. Balancing privacy and performance in federated learning: a systematic literature review on methods and metrics.Journal of Parallel and Distributed Computing, page 104918, 2024

work page 2024

[4] [4]

Client selection for federated learning with heterogeneous resources in mobile edge

Takayuki Nishio and Ryo Yonetani. Client selection for federated learning with heterogeneous resources in mobile edge. InICC 2019-2019 IEEE international conference on communications (ICC), pages 1–7. IEEE, 2019

work page 2019

[5] [5]

Federated learning with buffered asynchronous aggregation

John Nguyen, Kshitiz Malik, Hongyuan Zhan, Ashkan Yousefpour, Mike Rabbat, Mani Malek, and Dzmitry Huba. Federated learning with buffered asynchronous aggregation. InInternational Conference on Artificial Intelligence and Statistics, pages 3581–3607. PMLR, 2022

work page 2022

[6] [6]

Tifl: A tier-based federated learning system

Zheng Chai, Ahsan Ali, Syed Zawad, Stacey Truex, Ali Anwar, Nathalie Baracaldo, Yi Zhou, Heiko Ludwig, Feng Yan, and Yue Cheng. Tifl: A tier-based federated learning system. InProceedings of the 29th international symposium on high-performance parallel and distributed computing, pages 125–136, 2020. 14

work page 2020

[7] [7]

Asynchronous federated optimization.arXiv preprint arXiv:1903.03934, 2019

Cong Xie, Sanmi Koyejo, and Indranil Gupta. Asynchronous federated optimization.arXiv preprint arXiv:1903.03934, 2019

arXiv 1903

[8] [8]

Local differential privacy for deep learning.IEEE Internet of Things Journal, 7(7):5827– 5842, 2019

Pathum Chamikara Mahawaga Arachchige, Peter Bertok, Ibrahim Khalil, Dongxi Liu, Seyit Camtepe, and Mo- hammed Atiquzzaman. Local differential privacy for deep learning.IEEE Internet of Things Journal, 7(7):5827– 5842, 2019

work page 2019