EPIC: Efficient Private Image Classification (or: Learning from the Masters)

Makri, Eleftheria; Rotaru, Dragos; Smart, Nigel P.; Vercauteren, Fréderik

doi:10.1007/978-3-030-12612-4_24

Cited by 48 publications

(36 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The work also provides a simple easy-to-use API with a translator from Keras [51] to XONN. EPIC [52] demonstrates the use of transfer learning in the space of privacy-preserving machine learning while Quotient [53] takes the first steps in developing two party secure computation protocols for optimizers and normalizations. CrypTFlow [13] builds on SecureNN and uses trusted hardware to achieve maliciously secure protocols in a 3PC model.…”

Section: Privatementioning

confidence: 99%

Falcon: Honest-Majority Maliciously Secure Framework for Private Deep Learning

Wagh

Tople

Benhamouda³

et al. 2020

Proceedings on Privacy Enhancing Technologies

111

150

View full text Add to dashboard Cite

We propose Falcon, an end-to-end 3-party protocol for efficient private training and inference of large machine learning models. Falcon presents four main advantages – (i) It is highly expressive with support for high capacity networks such as VGG16 (ii) it supports batch normalization which is important for training complex networks such as AlexNet (iii) Falcon guarantees security with abort against malicious adversaries, assuming an honest majority (iv) Lastly, Falcon presents new theoretical insights for protocol design that make it highly efficient and allow it to outperform existing secure deep learning solutions. Compared to prior art for private inference, we are about 8× faster than SecureNN (PETS’19) on average and comparable to ABY3 (CCS’18). We are about 16 − 200× more communication efficient than either of these. For private training, we are about 6× faster than SecureNN, 4.4× faster than ABY3 and about 2−60× more communication efficient. Our experiments in the WAN setting show that over large networks and datasets, compute operations dominate the overall latency of MPC, as opposed to the communication.

show abstract

Section: Privatementioning

confidence: 99%

Falcon: Honest-Majority Maliciously Secure Framework for Private Deep Learning

Wagh

Tople

Benhamouda³

et al. 2020

Proceedings on Privacy Enhancing Technologies

111

150

View full text Add to dashboard Cite

show abstract

“…P3 abort if the received values are inconsistent. 3 ) in the boolean world, if we can get the garbled shares of x = (m v ⊕λ v,1 ) and y = (λ v,2 ⊕λ v,3 ), parties can use the free XOR technique to compute the garbled shares of v locally. Each of x, y is possessed by two parties, enabling them to verifiably generate the garbled shares using the protocol Π G vSh .…”

Section: Offlinementioning

confidence: 99%

“…Proof: To see the correctness, it suffices to show that 3 ). Note that r = 2 d r t + r d where r t denoted the truncated value of r and r d denoted the last d bits of r. Then, Cost Comparison:…”

Section: Convmentioning

confidence: 99%

“…Boolean to Garbled Sharing. d) Arithmetic to Garbled Sharing (A2G): Similar toΠ B2G , v = (m v − λ v,1 ) − (λ v,2 + λ v,3 ) and the parties can verifiably generate the garbled shares of x = (m v − λ v,1 ) and y = (λ v,2 + λ v,3 ) using Π G vSh . In the online phase, the parties evaluate a garbled subtractor circuit to obtain the shares of v = x − y.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Lessons Learned from SunDEW: A Self Defense Environment for Web Applications

Şahin¹,

Hebert²,

Oliveira³

2020

Proceedings 2020 Workshop on Measurements, Attacks, and Defenses for the Web

View full text Add to dashboard Cite

Machine learning has started to be deployed in fields such as healthcare and finance, which involves dealing with a lot of sensitive data. This propelled the need for and growth of privacy-preserving machine learning (PPML). We propose an actively secure four-party protocol (4PC), and a framework for PPML, showcasing its applications on four of the most widely-known machine learning algorithms -Linear Regression, Logistic Regression, Neural Networks, and Convolutional Neural Networks. (Communication). Protocol Π BitExt (Fig. 19) requires 1 round and a communication of 4 + 1 bits in the offline phase, while it requires 3 rounds and a communication of 5 + 2 bits in the online phase.Proof: During the offline phase, parties execute one instance each of Π vSh and Π B vSh resulting in one round and a communication of + 1 bits (Lemma C.1). Also, the offline phase for multiplication is performed resulting in an additional communication of 3 bits.During the online phase, parties first execute an arithmetic multiplication, resulting in one round and communication of 3 bits (Lemma B.4). The value rv is reconstructed towards both P 0 and P 3 , resulting in an additional round and an amortized communication of 2 bits. This is followed by the last round, where parties execute one instance of Π B vSh resulting in a communication of 2 bits. Thus the online phase requires three rounds and an amortized communication of 5 + 2 bits.Lemma D.4 (Communication). Protocol ReLU (Π relu ) requires 3 rounds and a communication of 8 + 2 bits in the offline phase, while it requires 4 rounds and a communication of 8 +2 bits in the online phase.

show abstract

“…Existing literature work has already shown that machine learning inferencing, upon which many fall detection systems rely, can be performed efficiently using MPC. This includes MPCbased image recognition using support vector machines (SVMs) [28] and remote rehabilitation treatment classification from patient data using decision trees [11].…”

Section: Introductionmentioning

confidence: 99%

Privacy-Enhancing Fall Detection from Remote Sensor Data Using Multi-Party Computation

Mainali¹,

Shepherd²

2019

Proceedings of the 14th International Conference on Availability, Reliability and Security

View full text Add to dashboard Cite

Motion-based fall detection systems are concerned with detecting falls from vulnerable users, which is typically performed by classifying measurements from a body-worn inertial measurement unit (IMU) using machine learning. Such systems, however, necessitate the collection of high-resolution measurements that may violate users' privacy, such as revealing their gait, activities of daily living (ADLs), and relative position using dead reckoning. In this paper, we investigate the application of multi-party computation (MPC) to IMU-based fall detection for protecting device measurement confidentiality. Our system is evaluated in a cloud-based setting that precludes parties from learning the underlying data using multiple, disparate cloud instances deployed in three geographical configurations. Using a publicly-available dataset, we demonstrate that MPC-based fall detection from IMU measurements is practical while achieving state-of-the-art error rates. In the best case, our system executes in 365.2 milliseconds, which falls well within the required time window for on-device data acquisition (750ms). CCS CONCEPTS• Security and privacy → Security services; Distributed systems security; Domain-specific security and privacy architectures; • Computer systems organization → Cloud computing.

show abstract

EPIC: Efficient Private Image Classification (or: Learning from the Masters)

Cited by 48 publications

References 36 publications

Falcon: Honest-Majority Maliciously Secure Framework for Private Deep Learning

Falcon: Honest-Majority Maliciously Secure Framework for Private Deep Learning

Lessons Learned from SunDEW: A Self Defense Environment for Web Applications

Privacy-Enhancing Fall Detection from Remote Sensor Data Using Multi-Party Computation

Contact Info

Product

Resources

About