When Secure Aggregation Falls Short: Achieving Long-Term Privacy in Asynchronous Federated Learning for LEO Satellite Networks

Mohamed Elmahallawy; Tie Luo

REVIEW 3 major objections 3 minor 37 references

When Secure Aggregation Falls Short: Achieving Long-Term Privacy in Asynchronous Federated Learning for LEO Satellite Networks

T0 review · 3 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read LTP-FLEO claims cumulative privacy leakage in asynchronous federated learning stays bounded by a single round, even over many training rounds.

desk verdict Useful, honestly argued paper with a real novelty—fixed-membership aggregation for long-term privacy—but the core guarantee hinges on an under-specified dropout fallback that the paper never stress-tests. read the letter →

arxiv 2508.13425 v1 pith:KQROFG73 submitted 2025-08-19 cs.CR

Mohamed Elmahallawy , Tie Luo This is my paper · ORCID

classification cs.CR

keywords federatedlearningsecureaggregationlong-termprivacyLEOsatellitenetworksasynchronouspartitioningmodelagebalancingfair

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 traditional secure aggregation fails in LEO satellite federated learning because satellites are only intermittently visible and because privacy is considered per round rather than over many rounds. It proposes LTP-FLEO, an asynchronous framework that groups satellites into partitions with synchronized visibility windows, refreshes secure-aggregation masks each round, and balances model age and contribution fairness. If its analysis is right, the cumulative privacy leakage over any number of rounds stays bounded by the leakage of a single round, so the server or an eavesdropper cannot piece together an individual satellite's updates by comparing aggregates over time. The framework also claims faster convergence, fairer aggregation, and model accuracy competitive with synchronous FL.

What carries the argument

The load-bearing mechanism is the combination of a fixed, visibility-synchronized partition of satellites and fresh secure-aggregation masks each round. Secure aggregation is a cryptographic protocol that lets the server sum client updates without seeing individual updates. Fixed membership prevents cross-round differencing attacks—the server always sees aggregates over the same set of clients, so each client's contribution is hidden by the same masking structure and cannot be isolated by subtracting aggregates from different rounds. The joint-visibility requirement is what makes the fixed partition feasible: satellites are grouped so that all members are simultaneously visible to the server

What would settle it

Run the protocol for many rounds in simulation with one partition member forced to miss its visibility window once. If the recovery path reveals that member's mask, then check whether the server can reconstruct that member's gradient by subtracting the aggregate from the previous round's aggregate; if the reconstructed gradient leaks more than the single-round bound, the non-compounding privacy claim fails. Alternatively, check whether reusing masks across rounds (a bug the paper warns against) causes leakage to grow linearly with $T$.

Watch

Extended reading notes

Core claim

The central discovery is a protocol structure that decouples per-round privacy from the number of rounds. In LTP-FLEO, satellites are partitioned by their predictable visibility so that every member of a partition is jointly visible to the server for each aggregation; within a partition, secure aggregation masks are freshly drawn each round and the membership is fixed. Because the same set of clients aggregates together every time, the server cannot form new linear combinations across rounds that isolate an individual update, and the total leakage over $T$ rounds remains bounded by the single-round secure-aggregation leakage rather than growing with $T$. The paper also introduces model age b

Load-bearing premise

The guarantee holds only if every satellite in a partition actually participates in every aggregation within the joint visibility window; if one member is missing, the protocol must either skip the round or expose that member's update, and the paper's analysis does not stress-test that failure mode.

Editorial extensions

If this is right

If the long-term privacy guarantee holds, asynchronous FL in LEO networks can be used for sensitive applications without worrying that the server or an eavesdropper will reconstruct individual updates by accumulating aggregates over time.
The privacy guarantee lets multi-round training proceed with the same per-round security budget, so the system does not need to increase mask sizes or add noise as training continues.
Fair aggregation based on contribution and model age balancing can reduce the negative effect of stale updates from short-visibility satellites, making convergence faster and fairer.
The partitioning strategy only requires predictable visibility, so the same protocol design could be adapted to other scheduled or orbital systems.
The competitive accuracy claim suggests the privacy-preserving asynchronous approach does not force a large accuracy penalty relative to synchronous FL.

Reading between the lines

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

The fixed-partition plus fresh-mask recipe is not tied to LEO specifically; any FL setting with predictable client availability, such as fleets of delivery vehicles or IoT sensors with duty cycles, could use the same structure if the joint-participation condition can be engineered.
A stronger adversary who can selectively knock out one partition member just before an aggregation could break the non-compounding guarantee; extending the analysis to Byzantine or drop-out-forcing adversaries would be a natural next test.
A concrete experiment: fix a partition of two satellites and two rounds, record the aggregate sums, and check whether the server can recover either satellite's update from the two aggregates; the paper's claim is that it cannot, which is directly testable.
The paper's convergence result carries a non-vanishing error floor that inherits the bounded-staleness assumption; an empirical sweep over staleness bounds would show how the accuracy-fairness trade-off behaves outside that assumption.

Share X Bluesky LinkedIn Reddit HN

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

The genuinely new piece here is the fixed-membership insight: if satellites are partitioned into jointly-visible groups that always aggregate together, then the server never sees a varying client set across rounds, so multi-round differencing can't isolate an individual update. That's a clean structural argument and it's presented clearly. The paper also does the right thing by stating its assumptions up front—fresh masks, no collusion, joint visibility—so you know exactly what the guarantee depends on.

What it does well: the formal privacy theorem is credible under those assumptions, the convergence analysis is honestly bounded (including the non-vanishing error floor), and the simulations seem to match the claims. The authors also situate the work against FedAsync and their own FEDLEO line, so the delta is roughly clear even if the grouping component is incremental.

The soft spot is real and load-bearing. The long-term privacy result only holds if every satellite in a partition actually submits during every aggregation window. In LEO, that's not a rare event; feeders drop, orbits shift, hardware glitches. The paper doesn't specify a protocol-level fallback for a missed window. If you aggregate without the missing satellite, the client set changes and the differencing attack comes back—masks being fresh doesn't help because the fixed-membership property is what blocked the attack. If you wait, you break the bounded-staleness assumptions behind the convergence theorem. The empirical section appears to only exercise full participation, so this failure mode is untested. That's exactly the kind of gap a referee should push on.

Minor concerns: the fairness result largely follows from defining weights proportional to contributions, and the partition optimizer is a heuristic without convergence guarantees. Neither is fatal, but they temper the scope.

Bottom line: this is a solid, honest paper that deserves serious peer review. It needs a fallback strategy for missed visibility windows—or at least a clear statement of the regime where the guarantee holds—before the headline claim is fully supported. I'd bring it to a reading group and would cite it if I worked in FL privacy for dynamic networks, even as a conditional result.

Referee Report

3 major / 3 minor

Summary. The paper proposes LTP-FLEO, an asynchronous federated learning framework for LEO satellite networks. Its main contribution is a privacy-aware satellite partitioning scheme that groups satellites with overlapping visibility into fixed partitions and enforces joint participation, combined with secure aggregation using fresh masks each round. The authors claim that this design yields long-term privacy: cumulative leakage over T rounds remains bounded by the single-round secure aggregation leakage, because aggregation membership is fixed and masks are refreshed. The paper also introduces model age balancing and fair global aggregation, and reports theoretical convergence guarantees and simulations showing improved fairness and accuracy.

Significance. If the long-term privacy result is correct, it fills a genuine gap: prior secure aggregation schemes for FL typically consider per-round privacy only, while asynchronous LEO settings with intermittent connectivity make multi-round differencing attacks a real threat. The central structural insight—that fixed aggregation sets plus fresh masks prevent cross-round composition—is credible under the stated assumptions. The paper also provides a concrete protocol for satellite partitioning and a convergence analysis, backed by simulations. The contribution is likely to be of interest to the FL and satellite networking communities, provided the protocol's behavior under missed visibility windows is rigorously specified and analyzed.

major comments (3)

[§4.2, Definition 2; §5.2, long-term privacy theorem] The long-term privacy guarantee depends critically on every satellite in a fixed partition submitting its update in every aggregation window (joint participation). The protocol groups satellites with overlapping visibility, but it does not specify a protocol-level fallback when a member misses its visibility window—a realistic event due to feeder-link outages, orbital perturbations, or hardware transients. If the server aggregates without the missing member, the set of participants changes; the server can then subtract the new aggregate from the previous one and isolate the missing satellite's masked update, collapsing the T-independent leakage bound. If the server waits, the bounded-staleness assumptions in §5.1 may be violated. The manuscript should specify the exact behavior in this failure case, state the resulting privacy/convergence guarantees, or modify the scheme to make the fail
[§4.2, joint participation enforcement; §5.2, Theorem conditions] The theorem's statement should explicitly list 'every partition member submits in every round' as an assumption, not an implicit consequence of the partitioning algorithm. The paper's use of predictable visibility makes joint participation plausible but not deterministic. In a LEO constellation, visibility windows are probabilistic; the analysis should provide a bound on the probability that all members submit, or discuss how the protocol detects and handles missed participation without leaking information. Without this, the headline claim 'long-term privacy across multi-round training' overstates the guarantee.
[§5.1, Assumptions 1–4; convergence theorem] The convergence analysis rests on bounded staleness, but the protocol's model age balancing and the possible need to delay aggregation until all partition members are available (if that fallback is chosen) create a tension. The paper does not prove that the age-balancing mechanism keeps staleness within the assumed bounds under realistic visibility patterns. The non-vanishing error floor is acknowledged, but the manuscript should clarify whether the convergence guarantee is conditional on the same joint-participation assumption as privacy, and quantify the maximum delay introduced by waiting for partition members. Otherwise, the 'accelerates global convergence' claim is not fully supported.

minor comments (3)

[Abstract] The phrase 'effectively safeguards both model and data privacy' is vague. Consider quantifying the leakage bound in terms of the minimum partition size k and the number of rounds T, and stating the security parameter explicitly.
[Algorithm 1] The greedy partition algorithm's near-optimality claim is asserted without a formal statement or proof. If this is used to justify the feasibility of the overall scheme, add a lemma or a reference to a standard approximation result.
[Notation] The distinction between 'round' and 'aggregation window' is sometimes blurred. Define both clearly at first use and use consistent subscripts in the equations of §5.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the long-term privacy claim is a derived property of the protocol, not an input restatement.

full rationale

Based on the available abstract (the only paper text provided), no circular step can be exhibited. The abstract describes a protocol design—fixed satellite partitions, enforced joint participation, and fresh secure-aggregation masks per round—and claims that this yields long-term privacy across rounds. That is a theorem about a construction, not a reduction of the conclusion to an input. The reader's concern that the privacy guarantee depends on every partition member participating in every aggregation is a robustness/assumption problem, not circularity: a conditional theorem does not define its own conclusion. Similarly, the leakage bound depending on partition size is standard secure-aggregation behavior, not a fitted parameter renamed as a prediction. No self-citation, imported uniqueness theorem, or ansatz smuggled via citation appears in the abstract. Therefore, without further textual evidence, the appropriate finding is no significant circularity.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The framework imports standard SecAgg guarantees and standard async-FL convergence assumptions; the paper's own design choices are the staleness exponent, partition size, and minimum joint visibility, all hand-set. No new physical or cryptographic entities are postulated. The headline long-term privacy result is therefore mostly a structural consequence of the fixed-partition design plus the imported SecAgg guarantee.

free parameters (3)

alpha (age-balancing exponent) = not stated; hand-set in (0,1]
Controls the staleness weight (1+tau)^(-alpha) in Section 4.3; it sets the accuracy/convergence tradeoff and is a design choice, not derived.
k (minimum partition size) = operator choice, e.g., k >= 2
The leakage bound is O(1/k); larger k improves privacy but constrains scheduling. This hand-chosen knob directly sets the headline privacy level.
L_min (minimum joint visibility duration) = not specified
Required for a partition to aggregate; determines which partitionings are feasible in the Section 4.2 optimization and is a design parameter.

assumptions (5)

domain assumption Satellite visibility is periodic and perfectly predictable from ephemeris (Section 3.1)
The fixed partition schedule is computed once from predicted visibility; any prediction error breaks joint participation and thus the privacy guarantee.
domain assumption Adversaries are honest-but-curious and do not collude with clients (Section 2.4 threat model)
The leakage bound does not cover a malicious server that corrupts clients or reveals masks.
domain assumption SecAgg masks are freshly generated per round and never revealed; at least k clients participate (Section 4.5)
The per-round information-theoretic guarantee of SecAgg is imported as a black box; if masks leak, the multi-round claim fails.
standard math Objective functions are L-smooth with bounded variance and bounded staleness (Section 5.1, Assumptions 1-4)
Standard assumptions in asynchronous FL convergence proofs; they produce the non-vanishing error floor in the convergence bound.
ad hoc to paper The greedy partition algorithm (Algorithm 1) produces a near-optimal feasible partition
The partition problem is claimed hard and solved heuristically; the approximation gap is not quantified, so achievable partitions may be suboptimal.

how reviews work

0 comments

Cite this review

Pith. "Pith review of When Secure Aggregation Falls Short: Achieving Long-Term Privacy in Asynchronous Federated Learning for LEO Satellite Networks." pith.science (2026). https://pith.science/paper/KQROFG73

@misc{pith2026250813425,
  author       = {Pith},
  title        = {Pith review of: When Secure Aggregation Falls Short: Achieving Long-Term Privacy in Asynchronous Federated Learning for LEO Satellite Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KQROFG73}},
  note         = {Machine review of arXiv:2508.13425}
}

read the original abstract

Secure aggregation is a common technique in federated learning (FL) for protecting data privacy from both curious internal entities (clients or server) and external adversaries (eavesdroppers). However, in dynamic and resource-constrained environments such as low Earth orbit (LEO) satellite networks, traditional secure aggregation methods fall short in two aspects: (1) they assume continuous client availability while LEO satellite visibility is intermittent and irregular; (2) they consider privacy in each communication round but have overlooked the possible privacy leakage through multiple rounds. To address these limitations, we propose LTP-FLEO, an asynchronous FL framework that preserves long-term privacy (LTP) for LEO satellite networks. LTP-FLEO introduces (i) privacy-aware satellite partitioning, which groups satellites based on their predictable visibility to the server and enforces joint participation; (ii) model age balancing, which mitigates the adverse impact of stale model updates; and (iii) fair global aggregation, which treats satellites of different visibility durations in an equitable manner. Theoretical analysis and empirical validation demonstrate that LTP-FLEO effectively safeguards both model and data privacy across multi-round training, promotes fairness in line with satellite contributions, accelerates global convergence, and achieves competitive model accuracy.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 36 canonical work pages

[1]

I. Ali, N. Al-Dhahir, and J. E. Hershey. Predicting the visibility of leo satellites. IEEE Transactions on Aerospace and Electronic Systems, 35 0 (4): 0 1183--1190, 1999

work page 1999
[2]

Ansys STK - Systems Tool Kit , 2024

Ansys Inc. Ansys STK - Systems Tool Kit , 2024. URL https://www.ansys.com/products/missions/ansys-stk. Accessed: 2025-04-14

work page 2024
[3]

Bonawitz, V

K. Bonawitz, V. Ivanov, B. Kreuter, A. Marcedone, H. B. McMahan, S. Patel, D. Ramage, A. Segal, and K. Seth. Practical secure aggregation for privacy-preserving machine learning. In proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security, pages 1175--1191, 2017

work page 2017
[4]

Chang, K

Y. Chang, K. Zhang, J. Gong, and H. Qian. Privacy-preserving federated learning via functional encryption, revisited. IEEE Transactions on Information Forensics and Security, 18: 0 1855--1869, 2023

work page 2023
[5]

Chen et al

W.-N. Chen et al. The fundamental price of secure aggregation in differentially private federated learning. In International Conference on Machine Learning, pages 3056--3089. PMLR, 2022 a

work page 2022
[6]

Y. Chen, M. Zhang, X. Li, T. Che, R. Jin, J. Guo, W. Yang, B. An, and X. Nie. Satellite-enabled internet of remote things network transmits field data from the most remote areas of the tibetan plateau. Sensors, 22 0 (10): 0 3713, 2022 b . doi:10.3390/s22103713

work page doi:10.3390/s22103713 2022
[7]

L. Deng. The mnist database of handwritten digit images for machine learning research. IEEE Signal Processing Magazine, 29 0 (6): 0 141--142, 2012

work page 2012
[8]

Y. Dong, C. Xiaojun, W. Jing, L. Kaiyun, and W. Wang. Meteor: improved secure 3-party neural network inference with reducing online communication costs. In Proceedings of the ACM Web Conference 2023, pages 2087--2098, 2023

work page 2023

Show all 37 references

[9]

Elmahallawy and T

M. Elmahallawy and T. Luo. AsyncFLEO : Asynchronous federated learning for LEO satellite constellations with high-altitude platforms. In 2022 IEEE International Conference on Big Data (Big Data), pages 5478--5487. IEEE, 2022

2022
[10]

Elmahallawy, T

M. Elmahallawy, T. Luo, and M. I. Ibrahem. Secure and efficient federated learning in LEO constellations using decentralized key generation and on-orbit model aggregation. In IEEE Global Communications Conference (GLOBECOM), 2023

2023
[11]

Elmahallawy, T

M. Elmahallawy, T. Luo, and K. Ramadan. Communication-efficient federated learning for leo satellite networks integrated with haps using hybrid noma-ofdm. IEEE Journal on Selected Areas in Communications, pages 1--1, 2024

2024
[12]

Q. Fang, Z. Zhai, S. Yu, Q. Wu, X. Gong, and X. Chen. Olive branch learning: A topology-aware federated learning framework for space-air-ground integrated network. IEEE Transactions on Wireless Communications, 22 0 (7): 0 4534--51, July 2023

2023
[13]

Geiping et al

J. Geiping et al. Inverting gradients-how easy is it to break privacy in federated learning? Advances in Neural Information Processing Systems (NeurIPS), 33: 0 16937--16947, 2020

2020
[14]

S. S. Hassan et al. Sfl-leo: Secure federated learning computation based on leo satellites for 6g non-terrestrial networks. In NOMS 2023-2023 IEEE/IFIP Network Operations and Management Symposium, pages 1--5. IEEE, 2023

2023
[15]

Helber, B

P. Helber, B. Bischke, A. Dengel, and D. Borth. Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 12 0 (7): 0 2217--2226, 2019

2019
[16]

Kanagavelu et al

R. Kanagavelu et al. Two-phase multi-party computation enabled privacy-preserving federated learning. In 2020 20th IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing (CCGRID), pages 410--419. IEEE, 2020

2020
[17]

Krizhevsky, V

A. Krizhevsky, V. Nair, and G. Hinton. Cifar-10 (canadian institute for advanced research). URL http://www. cs. toronto. edu/kriz/cifar. html, 5 0 (4): 0 1, 2010

2010
[18]

Q. Li, B. He, and D. Song. Practical one-shot federated learning for cross-silo setting. In Int. Joint Conf. on Artificial Intelligence (IJCAI), 2021

2021
[19]

X. Li, K. Huang, W. Yang, S. Wang, and Z. Zhang. On the convergence of fedavg on non-iid data. arXiv preprint arXiv:1907.02189, 2019

1907 arXiv
[20]

X. Li, K. Huang, W. Yang, S. Wang, and Z. Zhang. On the convergence of fedavg on non-iid data. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id=HJxNAnVtDS

2020
[21]

Z. Lin, Z. Chen, Z. Fang, X. Chen, X. Wang, and Y. Gao. Fedsn: A general federated learning framework over LEO satellite networks. arXiv preprint arXiv:2311.01483, 2023

2023 arXiv
[22]

Liu, H.-Y

Z. Liu, H.-Y. Lin, and Y. Liu. Long-term privacy-preserving aggregation with user-dynamics for federated learning. IEEE Transactions on Information Forensics and Security, 2023

2023
[23]

Ma, S.-A

J. Ma, S.-A. Naas, S. Sigg, and X. Lyu. Privacy-preserving federated learning based on multi-key homomorphic encryption. International Journal of Intelligent Systems, 37 0 (9): 0 5880--5901, 2022

2022
[24]

McMahan, E

B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas. Communication-efficient learning of deep networks from decentralized data. In AISTATS, pages 1273--1282, 2017

2017
[25]

M. Nasr, R. Shokri, and A. Houmansadr. Comprehensive privacy analysis of deep learning: Passive and active white-box inference attacks against centralized and federated learning. In 2019 IEEE symposium on security and privacy (SP), pages 739--753. IEEE, 2019

2019
[26]

\"O stman, P

J. \"O stman, P. Gomez, V. M. Shreenath, and G. Meoni. Decentralised semi-supervised onboard learning for scene classification in low-earth orbit. arXiv preprint arXiv:2305.04059, 2023

2023 arXiv
[27]

Ribero, H

M. Ribero, H. Vikalo, and G. De Veciana. Federated learning under intermittent client availability and time-varying communication constraints. IEEE Journal of Selected Topics in Signal Processing, 17 0 (1): 0 98--111, 2022

2022
[28]

F. D. Rossi, P. S. S. De Souza, and M. C. Luizelli. Resource allocation on low-earth orbit edge infrastructure: Taxonomy, survey, and research challenges. IEEE Access, 2025

2025
[29]

S. K. Routray, A. Javali, A. Sahoo, K. Sharmila, and S. Anand. Military applications of satellite based IoT . In 2020 Third International Conference on Smart Systems and Inventive Technology (ICSSIT), pages 122--127. IEEE, 2020

2020
[30]

Y. Shi, L. Zeng, J. Zhu, Y. Zhou, C. Jiang, and K. B. Letaief. Satellite federated edge learning: Architecture design and convergence analysis. IEEE Transactions on Wireless Communications, 2024

2024
[31]

J. So, R. E. Ali, B. G \"u ler, J. Jiao, and A. S. Avestimehr. Securing secure aggregation: Mitigating multi-round privacy leakage in federated learning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 9864--9873, 2023

2023
[32]

So et al

J. So et al. Secure aggregation for buffered asynchronous federated learning. In NeurIPS Workshop on New Frontiers in Federated Learning (NFFL), 2021

2021
[33]

K. Wei, J. Li, M. Ding, C. Ma, H. H. Yang, F. Farokhi, S. Jin, T. Q. Quek, and H. V. Poor. Federated learning with differential privacy: Algorithms and performance analysis. IEEE Transactions on Information Forensics and Security, 15: 0 3454--3469, 2020

2020
[34]

J. Wu, W. Zhang, and F. Luo. Esafl: Efficient secure additively homomorphic encryption for cross-silo federated learning. arXiv preprint arXiv:2305.08599, 2023

2023 arXiv
[35]

Xiang, S

M. Xiang, S. Ioannidis, E. Yeh, C. Joe-Wong, and L. Su. Efficient federated learning against heterogeneous and non-stationary client unavailability. Advances in Neural Information Processing Systems, 37: 0 104281--104328, 2024

2024
[36]

Xie et al

C. Xie et al. Asynchronous federated optimization. In 12 ^ th Wksp on Optimization for Machine Learning , 2020

2020
[37]

H. Yin, A. Mallya, A. Vahdat, J. M. Alvarez, J. Kautz, and P. Molchanov. See through gradients: Image batch recovery via gradinversion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16337--16346, 2021

2021

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

I. Ali, N. Al-Dhahir, and J. E. Hershey. Predicting the visibility of leo satellites. IEEE Transactions on Aerospace and Electronic Systems, 35 0 (4): 0 1183--1190, 1999

work page 1999

[2] [2]

Ansys STK - Systems Tool Kit , 2024

Ansys Inc. Ansys STK - Systems Tool Kit , 2024. URL https://www.ansys.com/products/missions/ansys-stk. Accessed: 2025-04-14

work page 2024

[3] [3]

Bonawitz, V

K. Bonawitz, V. Ivanov, B. Kreuter, A. Marcedone, H. B. McMahan, S. Patel, D. Ramage, A. Segal, and K. Seth. Practical secure aggregation for privacy-preserving machine learning. In proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security, pages 1175--1191, 2017

work page 2017

[4] [4]

Chang, K

Y. Chang, K. Zhang, J. Gong, and H. Qian. Privacy-preserving federated learning via functional encryption, revisited. IEEE Transactions on Information Forensics and Security, 18: 0 1855--1869, 2023

work page 2023

[5] [5]

Chen et al

W.-N. Chen et al. The fundamental price of secure aggregation in differentially private federated learning. In International Conference on Machine Learning, pages 3056--3089. PMLR, 2022 a

work page 2022

[6] [6]

Y. Chen, M. Zhang, X. Li, T. Che, R. Jin, J. Guo, W. Yang, B. An, and X. Nie. Satellite-enabled internet of remote things network transmits field data from the most remote areas of the tibetan plateau. Sensors, 22 0 (10): 0 3713, 2022 b . doi:10.3390/s22103713

work page doi:10.3390/s22103713 2022

[7] [7]

L. Deng. The mnist database of handwritten digit images for machine learning research. IEEE Signal Processing Magazine, 29 0 (6): 0 141--142, 2012

work page 2012

[8] [8]

Y. Dong, C. Xiaojun, W. Jing, L. Kaiyun, and W. Wang. Meteor: improved secure 3-party neural network inference with reducing online communication costs. In Proceedings of the ACM Web Conference 2023, pages 2087--2098, 2023

work page 2023