A performance analysis of VM-based Trusted Execution Environments for Confidential Federated Learning

Bruno Casella

REVIEW 4 major objections 5 minor 26 references

A performance analysis of VM-based Trusted Execution Environments for Confidential Federated Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read VM-based trusted execution environments like Intel TDX add no more than 1.5x overhead to confidential federated learning.

desk verdict Useful first TDX-for-FL benchmark, but the headline 1.5x cap is built on a single unrepeated measurement per cell. read the letter →

arxiv 2501.11558 v1 pith:2LDHA3JD submitted 2025-01-20 cs.CR cs.PF

Bruno Casella This is my paper · ORCID

classification cs.CRcs.PF

keywords federatedlearningconfidentialcomputingtrustedexecutionenvironmentIntelTDXSGXperformanceanalysisTLSoverheadOpenFL

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 claims that Intel TDX, a VM-based trusted execution environment, runs confidential federated learning with only a small, predictable runtime penalty: at most 1.5x relative to unprotected execution, compared with roughly 2x for an SGX-based pipeline. The evidence is a set of 100-round, one-epoch federated training runs across MNIST, CIFAR10, and CIFAR100 with ResNet-18 and MobileNetV3-Small, each measured under four configurations (baseline, baseline plus TLS, TDX, TDX plus TLS). The author decomposes the cost with an additive model in which TDX and TLS each contribute a fixed multiple of baseline time, with fitted coefficients of 0.126 and 0.245 for TDX on the two models. If the claim holds, confidential computing becomes a practical default for federated learning on public cloud and HPC resources, not a specialized option requiring heavy porting effort.

What carries the argument

The load-bearing mechanism is an additive runtime model, $T = T_{\text{baseline}} + O_{\text{TDX}} + O_{\text{TLS}}$, in which each security layer contributes an overhead $O = C \cdot T_{\text{baseline}}$ proportional to the unprotected baseline. The coefficients are fitted by non-negative linear regression and are the paper's quantitative result: $C_{\text{TDX}}$ is 0.126 for ResNet-18 and 0.245 for MobileNetV3-Small, while $C_{\text{TLS}}$ is 0.079 and 0.065 respectively. This identity matters because it separates the cost of computation isolation (TDX) from the cost of secure communication (TLS), so each can be measured, predicted, and budgeted independently. The hardware mechanism underneath is Intel TDX's VM-level isolation, which encrypts memory and CPU state while avoiding per-call enclave switching, and which is compared against an SGX-plus-Gramine baseline supplied by an earlier study.

What would settle it

Repeat the MobileNetV3-Small/MNIST 100-round training on the same 5th-generation Xeon hardware with several thread counts and multiple trials, comparing the best-tuned no-TEE baseline against TDX+TLS; the reported 78-minute vs 53-minute result is already 1.47x, so any trial pushing the ratio past 1.5x, or a better-tuned baseline, would refute the headline bound.

Watch

Extended reading notes

Core claim

The paper's central discovery is that VM-level isolation changes the cost profile of confidential federated learning. On the same 5th-generation Xeon hardware, TDX and TDX plus TLS completed the full training runs in at most 1.47x the unprotected baseline time, whereas the SGX-based pipeline took roughly twice as long. The overhead is additive and model-dependent: $T = T_{\text{baseline}} + C_{\text{TDX}}T_{\text{baseline}} + C_{\text{TLS}}T_{\text{baseline}}$ with $C_{\text{TDX}} = 0.126$ for ResNet-18 and $0.245$ for MobileNetV3-Small, and $C_{\text{TLS}} \approx 0.065\text{--}0.079$. The author attributes the improvement to TDX's VM-based design, which removes the constant enclave-to-unprotected-world switches that SGX requires, at the price of a larger trust boundary that includes the guest OS and VM administrators.

Load-bearing premise

The central claim assumes the TDX and baseline runs were configured under otherwise identical, representative conditions and that the single wall-clock measurements are stable; the paper does not report repeated runs or tuning checks, so a poorly tuned baseline could make the overhead appear smaller than it is.

Editorial extensions

If this is right

Secure federated training with TDX and TLS stays within 1.47x of unprotected training on every tested dataset/model pair, so the performance argument against confidential deployment on untrusted cloud or HPC hardware is substantially weakened.
Practitioners can estimate end-to-end runtime from a baseline measurement using the fitted coefficients ($C_{\text{TDX}} \approx 0.13\text{--}0.25$, $C_{\text{TLS}} \approx 0.07$), rather than needing to benchmark every security stack from scratch.
Because TDX removes the enclave-world switching and code-porting burden of SGX, VM-based TEEs make confidential FL compatible with standard federated learning implementations and workflows.
The roughly doubled TDX coefficient for MobileNetV3-Small implies that model architecture, especially layer count and depthwise separable convolutions, affects TEE overhead, so architecture-aware budgeting will matter.

Reading between the lines

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

Beyond the paper's comparisons, the TDX runs used 5th-generation Xeon Platinum 8592 CPUs while the SGX baseline used 3rd-generation Xeon Platinum 8380 CPUs, so the measured speed advantage probably mixes TEE architecture with hardware-generation gains; the paper does not isolate these factors.
The paper's linear model suggests a cheap extrapolation recipe—measure one secure run and estimate overhead for other models by subtracting the fitted coefficients—but this recipe is not validated across architectures and remains an unproven extension.
Since Table II reports single wall-clock values while Table III shows substantial variance in communication times, the headline 1.5x bound is likely run-dependent; repeated trials on the same hardware would reveal its distribution and true maximum.
A security-aware reading of the performance result: TDX's larger trust boundary means the low overhead is only acceptable when the guest OS and VM administrators are trusted, so the deployment recommendation should be conditioned on the threat model.

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

Quick take: this is a straightforward extension of the author's earlier SGX benchmark [17] to Intel TDX. The new data are the TDX wall-clock times for two models across three datasets, plus communication-time statistics. That is a legitimate contribution: TDX is new enough that public numbers help people planning confidential FL deployments. The paper does a clean job of describing the setup and reproducing the prior learning scenario, and the communication-time table with means and standard deviations is a nice touch.

The main claim, that TDX overhead is at most 1.5x, is consistent with the reported table but not actually established. The largest ratio is MobileNetV3-Small on MNIST: 1:18 vs 0:53, which is 1.47x. Those are single wall-clock runs reported at minute granularity, with no repetition or variance. A 2-3% margin is within ordinary run-to-run noise, so the 'at most 1.5x' phrasing is stronger than the evidence supports. The paper should either run repeats and report bounds, or soften the claim to 'around 1.5x or less in these runs.'

Other soft spots: the comparison to SGX runs on different CPU generations (5th Gen vs 3rd Gen Xeon), so the SGX-vs-TDX overhead comparison is not apples-to-apples. The paper also explicitly omits accuracy verification, saying the standard training routine makes it unnecessary. That is a defensible assumption but worth a sentence or two of justification or a quick accuracy table. The fitted linear model is descriptive, not predictive; that is fine as a summary but should not be called a predictive performance model. No code or data are provided, which limits reproducibility.

Who should read this: practitioners choosing between SGX and TDX for federated learning, and people benchmarking TEE overhead in general. It deserves peer review, but only conditionally: the single-run timing issue needs addressing, and the claims should be scaled back to match the evidence. The paper is on the right track and the data are genuinely new; with repeats and clearer limitations it would be a solid engineering data point.

I would send it to review, expecting revision.

Referee Report

4 major / 5 minor

Summary. The paper benchmarks Intel TDX, a VM-based TEE, for confidential federated learning under OpenFL, comparing wall-clock execution times with an earlier SGX/Gramine-based baseline from the same authors' prior work. Three datasets (MNIST, CIFAR10, CIFAR100) and two models (ResNet-18, MobileNetV3-Small) are trained for 100 federated rounds, with incremental protections: baseline, TLS, TDX, and TDX+TLS. The authors report that TDX adds only slight overhead, state in the abstract that VM-based TEEs introduce at most 1.5x overhead, and fit a linear overhead model T = T_baseline + C_TDX*T_baseline + C_TLS*T_baseline to the measured times. They conclude that confidential computing can be deployed in untrusted HPC/cloud environments without significant performance detriment.

Significance. If the central quantitative claim were well supported, the paper would provide useful deployment evidence for VM-based TEEs in federated learning, an area where most prior measurements concern SGX-style application isolation. The study has concrete strengths: it covers three datasets and two model architectures, uses a realistic distributed setting with one aggregator and three collaborators, and reports communication-time statistics in Table III. The linear overhead model is transparent and easy to reuse. However, the headline 'at most 1.5x' claim is currently underwritten by single rounded wall-clock measurements whose largest ratio is only 1.47x, and the fitted overhead coefficients are descriptive rather than predictive. The SGX comparison is also confounded by different CPU generations. These issues bear directly on the paper's main conclusion.

major comments (4)

[Abstract and Table II] The central claim that VM-based TEEs introduce 'at most 1.5x' overhead is not supported by the evidence as reported. The largest observed ratio in Table II is MobileNetV3-Small on MNIST with TDX+TLS: 1:18 versus 0:53 baseline, which is 1.47x. Each configuration is represented by a single wall-clock time rounded to the minute, with no repeated runs, standard deviations, or confidence intervals, leaving a margin of only about 2-3% below the asserted cap. Rounding alone could hide a true ratio above 1.5x, and run-to-run variance from network jitter, data loading, or VM scheduling could do the same. The abstract should either be qualified to the observed point estimates or the authors should provide repeated measurements with error bars demonstrating that the 1.5x bound holds across runs.
[Section III, overhead model and Fig. 2] The overhead coefficients C_TDX and C_TLS are fitted by non-negative linear regression to the same wall-clock data from which the overhead ratios are computed, yet no goodness-of-fit, residual analysis, or out-of-sample validation is provided. The model is therefore descriptive of the six measured configurations, not a predictive performance model. In addition, the model form is inherited from prior work [17] without revalidation for TDX. The manuscript should state this limitation explicitly and should not present the fitted coefficients as evidence that the 1.5x bound generalizes beyond the measured runs.
[Section III, experimental setup and Table II] The comparison between TDX and SGX is confounded by hardware differences: the TDX experiments run on dual-socket 5th-generation Intel Xeon Platinum 8592 processors, while the SGX baseline runs on dual-socket 3rd-generation Intel Xeon Platinum 8380 processors. The narrative that SGX 'doubles execution time' while TDX adds only slight overhead conflates TEE technology with CPU generation. The authors should either run both TEE configurations on the same hardware generation or restrict the SGX comparison to qualitative statements and make the hardware confound explicit.
[Section III, accuracy statement] The manuscript states: 'Since our approach is not affected by loss in learning performance, as we perform a standard training routine as in [17], and due to space constraints, we omit plots and discussion about model's accuracies.' This assertion is unsupported because no accuracy numbers are reported anywhere in the paper. While the paper is focused on performance, claiming no learning-performance loss without reporting accuracies is a load-bearing omission, especially because the sentence frames the omission as a consequence of the claim rather than as a measurement. The authors should either report the accuracy results or remove the claim.

minor comments (5)

[Section III, Figs. 1 and 2] The items labeled 'Fig. 1' and 'Fig. 2' are not figures but displayed equations and a coefficient table, respectively; they should be numbered as equations or tables and referenced consistently in the text.
[Header/abstract metadata] The template placeholder 'Index Terms—component, formatting, style, styling, insert' has been left in the manuscript and should be replaced with actual index terms.
[Section III, sentence on CIFAR/MNIST statistics] The parenthetical sentence about CIFAR100 and 'colours MNIST' is grammatically garbled and unclear; it should be rewritten, and the claim about MNIST being colored should be checked, as MNIST is grayscale.
[Section III, discussion of MobileNetV3] The sentence 'The author hypothesizes that this can be due to some optimization reasons' uses the singular 'the author' where 'we' or 'the authors' is intended; this should be corrected.
[Table II presentation] The two blocks of Table II (TDX-based rows and SGX-based rows) are not explicitly labeled in the table itself; adding a column or grouping label indicating the hardware/TEE type would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the at-most-1.5x overhead claim is a direct summary of measured wall-clock ratios, and the fitted linear model is descriptive rather than predictive.

full rationale

The paper's central quantitative claim, that VM-based TEEs introduce at most 1.5x overhead, is obtained directly from the wall-clock measurements in Table II by comparing TDX+TLS runs against the paper's own baseline runs (e.g., 1:18 / 0:53 = 1.47x for MobileNetV3-Small on MNIST). This is a direct measurement ratio, not the output of a fitted model or the consequence of a self-citation. The linear performance model T = Tbaseline + C_TDX*Tbaseline + C_TLS*Tbaseline is fit by non-negative regression to the same Table II data and is explicitly used 'to better understand the overhead added by each single mechanism' (Section III), not to predict a held-out or future quantity; no fitted value is renamed as a prediction. The paper inherits the SGX baseline and model structure from the author's prior work [17], which is a self-citation, but the absolute TDX overhead conclusion rests on the current paper's own baseline measurements. Concerns about single unrepeated runs, minute-level rounding, and different CPU generations for the SGX comparison are experimental-rigor and comparability issues, not circularity. The derivation chain is therefore self-contained with respect to its main claim.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central overhead bound is an empirical measurement, so the main underlying assumptions are about the stability and comparability of the benchmark runs. The fitted coefficients are free parameters, and the paper adds no invented entities.

free parameters (4)

C_TDX (ResNet-18) = 0.126
Fitted via non-negative linear regression to the wall-clock times of TDX runs relative to baseline; descriptive, not predictive.
C_TDX (MobileNetV3-Small) = 0.245
Fitted the same way; the paper notes the MobileNet coefficient is roughly double ResNet-18.
C_TLS (ResNet-18) = 0.079
Fitted TLS overhead coefficient.
C_TLS (MobileNetV3-Small) = 0.065
Fitted TLS overhead coefficient.

assumptions (3)

domain assumption Linear additive performance model T = T_baseline + C_TDX * T_baseline + C_TLS * T_baseline
The paper states 'we reproduced the performance model presented in [17]' without validating it on held-out data; the model form is assumed to hold for TDX.
domain assumption No accuracy loss from standard training
The paper says 'Since our approach is not affected by loss in learning performance ... we omit plots and discussion about model's accuracies' (Section III). This is asserted, not demonstrated.
domain assumption Single-run wall-clock times are representative
Table II reports one time per configuration with no repetitions or error bars, so the measured overheads are treated as stable.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A performance analysis of VM-based Trusted Execution Environments for Confidential Federated Learning." pith.science (2026). https://pith.science/paper/2LDHA3JD

@misc{pith2026250111558,
  author       = {Pith},
  title        = {Pith review of: A performance analysis of VM-based Trusted Execution Environments for Confidential Federated Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2LDHA3JD}},
  note         = {Machine review of arXiv:2501.11558}
}

read the original abstract

Federated Learning (FL) is a distributed machine learning approach that has emerged as an effective way to address recent privacy concerns. However, FL introduces the need for additional security measures as FL alone is still subject to vulnerabilities such as model and data poisoning and inference attacks. Confidential Computing (CC) is a paradigm that, by leveraging hardware-based trusted execution environments (TEEs), protects the confidentiality and integrity of ML models and data, thus resulting in a powerful ally of FL applications. Typical TEEs offer an application-isolation level but suffer many drawbacks, such as limited available memory and debugging and coding difficulties. The new generation of TEEs offers a virtual machine (VM)-based isolation level, thus reducing the porting effort for existing applications. In this work, we compare the performance of VM-based and application-isolation level TEEs for confidential FL (CFL) applications. In particular, we evaluate the impact of TEEs and additional security mechanisms such as TLS (for securing the communication channel). The results, obtained across three datasets and two deep learning models, demonstrate that the new VM-based TEEs introduce a limited overhead (at most 1.5x), thus paving the way to leverage public and untrusted computing environments, such as HPC facilities or public cloud, without detriment to performance.

Figures

Figures reproduced from arXiv: 2501.11558 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 16 canonical work pages

[17]

& Aldinucci, M

Casella, B., Colonnelli, I., Mittone, G., Birke, R., Ri viera, W., Scia- rappa, A., Cavazzoni, C. & Aldinucci, M. A Performance Analy sis for Conﬁdential Federated Learning. IEEE Security And Privacy, SP 2024 - W orkshops, San Francisco, CA, USA, May 23, 2024 . pp. 40-47 (2024), https://doi.org/10.1109/SPW63631.2024.00009

arXiv 2024
[1]

& Arcas, B

McMahan, B., Moore, E., Ramage, D., Hampson, S. & Arcas, B . Communication-Efﬁcient Learning of Deep Networks from Dec entral- ized Data. Proc. Of The 20th Intl. Conference On Artiﬁcial Intelligenc e And Statistics, AISTATS 2017, 20-22 April 2017, F ort Lauder dale, FL, USA. vol. 54 pp. 1273-1282 (2017)

work page 2017
[2]

& Roth, A

Dwork, C. & Roth, A. The Algorithmic Foundations of Diffe rential Privacy. F oundations And Trends® In Theoretical Computer Science . 9, 211-407 (2014), http://dx.doi.org/10.1561/0400000042

doi:10.1561/0400000042 2014
[3]

& Rosulek, M

Evans, D., Kolesnikov, V . & Rosulek, M. A Pragmatic Intro duction to Secure Multi-Party Computation. F ound. Trends Priv. Secur .. 2, 70-246 (2018), https://doi.org/10.1561/3300000019

doi:10.1561/3300000019 2018
[4]

& Scarlata, V

Anati, I., Gueron, S., Johnson, S. & Scarlata, V . Innovat ive Technology for CPU Based Attestation and Sealing. (2013), https://api.semanticscholar.org/CorpusID:14218854

work page 2013
[5]

TrustZone: Integrated hardware and s oftware secu- rity

Alves T., Felton D. TrustZone: Integrated hardware and s oftware secu- rity. Tech. In-Depth 3, 4 (2004), 18–24

work page 2004
[6]

& V aswani, K

Guo, J., Pietzuch, P ., Paverd, A. & V aswani, K. Trustwort hy AI Using Conﬁdential Federated Learning. Commun. ACM . 67, 48-53 (2024,8), https://doi.org/10.1145/3677390

doi:10.1145/3677390 2024
[7]

Attention is All you Need

V aswani, A., et al., I. Attention is All you Need. Advances In Neural Information Processing Systems 30: Annual Conference On Ne ural Information Processing Systems 2017, December 4-9, 2017, L ong Beach, CA, USA . pp. 5998-6008 (2017)

work page 2017

Show all 26 references

[8]

Llama 2: Open Foundation and Fine-Tuned Chat Models

Touvron, H., et al. Llama 2: Open Foundation and Fine-Tuned Chat Models. CoRR. abs/2307.09288 (2023), https://doi.org/10.48550/arXiv.2307.09288

2023 doi arXiv
[9]

OpenAI GPT-4 Technical Report. CoRR. abs/2303.08774 (2023), https://doi.org/10.48550/arXiv.2303.08774

2023 doi arXiv
[10]

Cross-Facility Federated Learn ing

Colonnelli, I., et al. Cross-Facility Federated Learn ing. Procedia Com- puter Science . 240 pp. 3-12 (2024), Proceedings of the First EuroHPC user day

2024
[11]

& Aldinucci, M

Colonnelli, I., Casella, B., Mittone, G., Arfat, Y ., Ca ntalupo, B., Espos- ito, R., Martinelli, A., Medi´ c, D. & Aldinucci, M. Federate d Learning Meets HPC and Cloud. Machine Learning F or Astrophysics. pp. 193-199 (2023)

2023
[12]

& Shmatikov, V

Shokri, R., Stronati, M., Song, C. & Shmatikov, V . Membe rship Infer- ence Attacks Against Machine Learning Models. 2017 IEEE Symposium On Security And Privacy, SP 2017, San Jose, CA, USA, May 22-26 ,

2017
[13]

& Ristenpart, T

Fredrikson, M., Jha, S. & Ristenpart, T. Model Inversio n Attacks that Exploit Conﬁdence Information and Basic Countermeasu res. Pro- ceedings Of The 22nd ACM SIGSAC Conference On Computer And Communications Security, Denver , CO, USA, October 12-16, 2 015. pp. 1322-1333 (2015...

2015 doi
[14]

& Borisov, N

Ganju, K., Wang, Q., Y ang, W., Gunter, C. & Borisov, N. Pr operty Inference Attacks on Fully Connected Neural Networks using Per- mutation Invariant Representations. Proceedings Of The 2018 ACM SIGSAC Conference On Computer And Communications Security , CCS 2018, Toronto, ON,...

2018
[15]

& Kourtel- lis, N

Mo, F., Haddadi, H., Katevas, K., Marin, E., Perino, D. & Kourtel- lis, N. PPFL: privacy-preserving federated learning with t rusted ex- ecution environments. MobiSys ’21: The 19th Annual International Conference On Mobile Systems, Applications, And Services, Virtual Event, Wi...

2021
[16]

& Fetzer, C

Quoc, D. & Fetzer, C. SecFL: Conﬁdential Federated Lear ning using TEEs. CoRR. abs/2110.00981 (2021), https://arxiv.org/abs/2110.00 981

2021 arXiv
[18]

Cooperation and security isolation of l ibrary OSes for multi-process applications

Tsai, C., et al. Cooperation and security isolation of l ibrary OSes for multi-process applications. Ninth Eurosys Conference 2014, EuroSys 2014, Amsterdam, The Netherlands, April 13-16, 2014 . pp. 9:1-9:14 (2014)

2014
[19]

SCONE: Secure Linux Containers wit h Intel SGX

Arnautov, S., et al. SCONE: Secure Linux Containers wit h Intel SGX. 12th USENIX Symposium On Operating Systems Design And Imple - mentation, OSDI 2016, Savannah, GA, USA, November 2-4, 2016 . pp. 689-703 (2016)

2016
[20]

OpenFL: the open federated learning li brary

Foley, P ., et al. OpenFL: the open federated learning li brary. Physics In Medicine & Biology . (2022)

2022
[21]

The Transport Layer Security (TLS) Protoc ol V ersion 1.3

Rescorla, E. The Transport Layer Security (TLS) Protoc ol V ersion 1.3. (RFC Editor,2018,8), https://www.rfc-editor.org/info/rfc8446

2018
[22]

& Haffner, P

LeCun, Y ., Bottou, L., Bengio, Y . & Haffner, P . Gradient -based learning applied to document recognition. Proc. IEEE . 86, 2278-2324 (1998)

1998
[23]

& Others Learning multiple layers of features from tiny images

Krizhevsky, A., Hinton, G. & Others Learning multiple layers of features from tiny images. (Citeseer,2009), https://www.cs.toronto.edu/ kriz/learning-features-2009-TR.pdf

2009
[24]

& Sun, J

He, K., Zhang, X., Ren, S. & Sun, J. Deep Residual Learnin g for Image Recognition. 2016 IEEE Conference On Computer Vision And Pattern Recognition, CVPR 2016, Las V egas, NV , USA, June 27-30, 2016 . pp. 770-778 (2016)

2016
[25]

fig1.png

Howard, A., et al. & Zhu, Y . Searching for MobileNetV3. 2019 IEEE/CVF International Conference On Computer Vision, ICC V 2019, Seoul, Korea (South), October 27 - November 2, 2019 . pp. 1314-1324 (2019), https://doi.org/10.1109/ICCV .2019.00140 This figure "fig1.png" is availa...

2019
[2017]

3-18 (2017), https://doi.org/10.1109/SP .2017.41

pp. 3-18 (2017), https://doi.org/10.1109/SP .2017.41

2017 doi

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [17]

& Aldinucci, M

Casella, B., Colonnelli, I., Mittone, G., Birke, R., Ri viera, W., Scia- rappa, A., Cavazzoni, C. & Aldinucci, M. A Performance Analy sis for Conﬁdential Federated Learning. IEEE Security And Privacy, SP 2024 - W orkshops, San Francisco, CA, USA, May 23, 2024 . pp. 40-47 (2024), https://doi.org/10.1109/SPW63631.2024.00009

arXiv 2024

[2] [1]

& Arcas, B

McMahan, B., Moore, E., Ramage, D., Hampson, S. & Arcas, B . Communication-Efﬁcient Learning of Deep Networks from Dec entral- ized Data. Proc. Of The 20th Intl. Conference On Artiﬁcial Intelligenc e And Statistics, AISTATS 2017, 20-22 April 2017, F ort Lauder dale, FL, USA. vol. 54 pp. 1273-1282 (2017)

work page 2017

[3] [2]

& Roth, A

Dwork, C. & Roth, A. The Algorithmic Foundations of Diffe rential Privacy. F oundations And Trends® In Theoretical Computer Science . 9, 211-407 (2014), http://dx.doi.org/10.1561/0400000042

doi:10.1561/0400000042 2014

[4] [3]

& Rosulek, M

Evans, D., Kolesnikov, V . & Rosulek, M. A Pragmatic Intro duction to Secure Multi-Party Computation. F ound. Trends Priv. Secur .. 2, 70-246 (2018), https://doi.org/10.1561/3300000019

doi:10.1561/3300000019 2018

[5] [4]

& Scarlata, V

Anati, I., Gueron, S., Johnson, S. & Scarlata, V . Innovat ive Technology for CPU Based Attestation and Sealing. (2013), https://api.semanticscholar.org/CorpusID:14218854

work page 2013

[6] [5]

TrustZone: Integrated hardware and s oftware secu- rity

Alves T., Felton D. TrustZone: Integrated hardware and s oftware secu- rity. Tech. In-Depth 3, 4 (2004), 18–24

work page 2004

[7] [6]

& V aswani, K

Guo, J., Pietzuch, P ., Paverd, A. & V aswani, K. Trustwort hy AI Using Conﬁdential Federated Learning. Commun. ACM . 67, 48-53 (2024,8), https://doi.org/10.1145/3677390

doi:10.1145/3677390 2024

[8] [7]

Attention is All you Need

V aswani, A., et al., I. Attention is All you Need. Advances In Neural Information Processing Systems 30: Annual Conference On Ne ural Information Processing Systems 2017, December 4-9, 2017, L ong Beach, CA, USA . pp. 5998-6008 (2017)

work page 2017