Occlumency

Lee, Taegyeong; Lin, Zhi-Qi; Pushp, Saumay; Li, Caihua; Liu, Yunxin; Lee, Mi Kyung; Xu, Fengyuan; Xu, Chenren; Zhang, Lintao; Song, Junehwa

doi:10.1145/3300061.3345447

Cited by 76 publications

(12 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Origami achieves 11× performance improvement than a pure SGX approach, and 15.1× performance improvement compared to Slalom [142]. Aiming to address the memory limitation and page swapping of SGX enclave, Lee et al [84] carefully developed on-demand weights loading, memory-efficient inference, and parallel processing pipelines in their proposed Occlumency, which gives a 3.6× speedup compared to a pure TEE-based method and 72% latency overhead compared to a pure GPU approach. In [125], Alexander et al presented the eNNclave tool-chain to cut TensorFlow models at any layers and split them into public and enclave layers, where GPU performs public layers for acceleration.…”

Section: Research Status Of Soi Due To the Similarity Of Computation ...mentioning

confidence: 99%

A Survey of Secure Computation Using Trusted Execution Environments

Li¹,

Yang²,

Xiang³

et al. 2023

Preprint

View full text Add to dashboard Cite

As an essential technology underpinning trusted computing, the trusted execution environment (TEE) allows one to launch computation tasks on both on-and off-premises data while assuring confidentiality and integrity. This article provides a systematic review and comparison of TEE-based secure computation protocols. We first propose a taxonomy that classifies secure computation protocols into three major categories, namely secure outsourced computation, secure distributed computation and secure multi-party computation. To enable a fair comparison of these protocols, we also present comprehensive assessment criteria with respect to four aspects: setting, methodology, security and performance. Based on these criteria, we review, discuss and compare the state-of-the-art TEE-based secure computation protocols for both general-purpose computation functions and special-purpose ones, such as privacy-preserving machine learning and encrypted database queries. To the best of our knowledge, this article is the first survey to review TEE-based secure computation protocols and the comprehensive comparison can serve as a guideline for selecting suitable protocols for deployment in practice. Finally, we also discuss several future research directions and challenges.CCS Concepts: • Security and privacy → Tamper-proof and tamper-resistant designs; Trust frameworks; Privacy-preserving protocols; Management and querying of encrypted data.

show abstract

Section: Research Status Of Soi Due To the Similarity Of Computation ...mentioning

confidence: 99%

A Survey of Secure Computation Using Trusted Execution Environments

Li¹,

Yang²,

Xiang³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Model Execution with Memory Budget. To accommodate tight memory budgets, two popular solutions are explored in the literature: model compression [10], [11], [12], [13], [14], [15], [16], [52] and offloading [17], [18], [19], [20], [21], [22]. Model compression techniques reduce the model size by removing redundant parameters such as layers, filters and channels [10], lowering the parameter precision [15], searching efficient model architectures [16], etc.…”

Section: Related Workmentioning

confidence: 99%

“…However, when a model is compressed, its accuracy or robustness is often compromised, which is not favorable in mission-critical applications, e.g., self-driving. Research on offloading dynamically changes model partition position [22], optimizes offloading patterns to reduce delay [53], improves the inference privacy [20], etc. Offloading does not harm model accuracy but requires network connections, making it vulnerable to network fluctuations.…”

Section: Related Workmentioning

confidence: 99%

“…Conventional solutions to efficient DNN inference with limited memory budget include model compression [10], [11], [12], [13], [14], [15], [16] and cloud offloading [17], [18], [19], [20], [21], [22]. Model compression techniques remove redundant parameters or decrease the precision of model parameters in the DNN, but usually induce accuracy losses at high compression rates, which is undesired in mission-critical applications, e.g., self-driving.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

DNN‐aided read‐voltage threshold optimization for MLC flash memory with finite block length

et al. 2021

View full text Add to dashboard Cite

The error‐correcting performance of multi‐level‐cell (MLC) NAND flash memory is closely related to the block length of error‐correcting codes (ECCs) and log‐likelihood‐ratios of the read‐voltage thresholds. Driven by this issue, this paper optimizes the read‐voltage thresholds for MLC flash memory to improve the decoding performance of ECCs with finite block length. First, through the analysis of channel coding rate and decoding error probability under finite block length, the optimization problem of read‐voltage thresholds to minimize the maximum decoding error probability is formulated. Second, a cross‐iterative search algorithm to optimize read‐voltage thresholds under the perfect knowledge of flash memory channel is developed. However, it is challenging to analytically characterize the voltage distribution under the effect of data retention noise. To address this problem, a deep neural network (DNN)‐aided optimization strategy to optimize the read‐voltage thresholds is developed, where a multi‐layer perception network is employed to learn the relationship between voltage distribution and read‐voltage thresholds. Simulation results show that, compared with the existing schemes, the proposed DNN‐aided read‐voltage threshold optimization strategy with a well‐designed Low Density Parity Check (LDPC) code can not only improve the program‐and‐erase endurance but also reduce the read latency.

show abstract

“…The inference latency has undoubtedly become the most severe obstacle for SGX to develop deep learning systems in the cloud [8,9]. Taegyeong Lee [10] found that deep learning inference inside an Enclave is up to 6.4 times slower than running outside the Enclave. The main reason for the performance degradation is the hardware design limitation.…”

Section: Introductionmentioning

confidence: 99%

Branchy-TEE: Deep Learning Security Inference Acceleration Using Trusted Execution Environment

Wang,

Deng,

Meng

et al. 2023

International Conferences on Software Engineering and Knowledge Engineering

View full text Add to dashboard Cite

Deep Learning as a Service (DLaaS) has become a remarkable trend in modern data-driven online services. Both data holders and service providers need to build on trust in thirdparty cloud infrastructure platforms. However, once the trust is broken, data holders' sensitive data and service providers' intellectual property rights will face significant security and privacy risks. In this paper, we propose a secure and efficient inference framework for deep learning in untrustworthy cloud platforms, termed Branchy-TEE, which aims to protect the confidentiality and integrity of data and models of multiple participating actors throughout the inference process using the Trusted Execution Environment (TEE). Branchy-TEE dynamically loads the inference network into the TEE on-demand based on early-exit mechanism, expecting to break the hardware performance bottleneck of the TEE. Moreover, a joint training method based on knowledge distillation for multi-exit networks is proposed, by flowing "knowledge" from the final exit with high accuracy to the early branch exit with lower accuracy. Finally, the effectiveness and efficiency of Branchy-TEE are verified through extensive experiments in real environments, while achieving an optimal balance between performance and hardware resources.

show abstract

Occlumency

Cited by 76 publications

References 16 publications

A Survey of Secure Computation Using Trusted Execution Environments

A Survey of Secure Computation Using Trusted Execution Environments

DNN‐aided read‐voltage threshold optimization for MLC flash memory with finite block length

Branchy-TEE: Deep Learning Security Inference Acceleration Using Trusted Execution Environment

Contact Info

Product

Resources

About