TY - GEN
T1 - Edge-assisted CNN inference over encrypted data for internet of things
AU - Tian, Yifan
AU - Yuan, Jiawei
AU - Yu, Shucheng
AU - Hou, Yantian
AU - Song, Houbing
N1 - Publisher Copyright:
© ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2019.
PY - 2019
Y1 - 2019
N2 - Supporting the inference tasks of convolutional neural network (CNN) on resource-constrained Internet of Things (IoT) devices in a timely manner has been an outstanding challenge for emerging smart systems. To mitigate the burden on IoT devices, one prevalent solution is to offload the CNN inference tasks to the public cloud. However, this “offloading-to-cloud” solution may cause privacy breach since the offloaded data can contain sensitive information. For privacy protection, the research community has resorted to advanced cryptographic primitives to support CNN inference over encrypted data. Nevertheless, these attempts are limited by the real-time performance due to the heavy IoT computational overhead brought by cryptographic primitives. In this paper, we propose an edge-computing-assisted scheme to boost the efficiency of CNN inference tasks on IoT devices, which also protects the privacy of IoT data to be offloaded. In our scheme, the most time-consuming convolutional and fully-connected layers are offloaded to edge computing devices and the IoT device only performs efficient encryption and decryption on the fly. As a result, our scheme enables IoT devices to securely offload over 99% CNN operations, and edge devices to execute CNN inference over encrypted data as efficiently as on plaintext. Experiments on AlexNet show that our scheme can speed up CNN inference for more than 35× with a 95.56% energy saving for IoT devices.
AB - Supporting the inference tasks of convolutional neural network (CNN) on resource-constrained Internet of Things (IoT) devices in a timely manner has been an outstanding challenge for emerging smart systems. To mitigate the burden on IoT devices, one prevalent solution is to offload the CNN inference tasks to the public cloud. However, this “offloading-to-cloud” solution may cause privacy breach since the offloaded data can contain sensitive information. For privacy protection, the research community has resorted to advanced cryptographic primitives to support CNN inference over encrypted data. Nevertheless, these attempts are limited by the real-time performance due to the heavy IoT computational overhead brought by cryptographic primitives. In this paper, we propose an edge-computing-assisted scheme to boost the efficiency of CNN inference tasks on IoT devices, which also protects the privacy of IoT data to be offloaded. In our scheme, the most time-consuming convolutional and fully-connected layers are offloaded to edge computing devices and the IoT device only performs efficient encryption and decryption on the fly. As a result, our scheme enables IoT devices to securely offload over 99% CNN operations, and edge devices to execute CNN inference over encrypted data as efficiently as on plaintext. Experiments on AlexNet show that our scheme can speed up CNN inference for more than 35× with a 95.56% energy saving for IoT devices.
KW - Convolutional neural network
KW - Deep learning
KW - Edge computing
KW - Internet of Things
KW - Privacy
UR - http://www.scopus.com/inward/record.url?scp=85077503485&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85077503485&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-37228-6_5
DO - 10.1007/978-3-030-37228-6_5
M3 - Conference contribution
AN - SCOPUS:85077503485
SN - 9783030372279
T3 - Lecture Notes of the Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering, LNICST
SP - 85
EP - 104
BT - Security and Privacy in Communication Networks - 15th EAI International Conference, SecureComm 2019, Proceedings
A2 - Chen, Songqing
A2 - Choo, Kim-Kwang Raymond
A2 - Fu, Xinwen
A2 - Lou, Wenjing
A2 - Mohaisen, Aziz
T2 - 15th International Conference on Security and Privacy in Communication Networks, SecureComm 2019
Y2 - 23 October 2019 through 25 October 2019
ER -