Knowledge Distillation for Mobile Edge Computation Offloading

doi:10.12142/ZTECOM.202002006

ZTE Communications ›› 2020, Vol. 18 ›› Issue (2): 40-48.DOI: 10.12142/ZTECOM.202002006

• Special Topic • Previous Articles Next Articles

Knowledge Distillation for Mobile Edge Computation Offloading

CHEN Haowei, ZENG Liekang, YU Shuai, CHEN Xu()

School of Data and Computer Science, Sun Yat?sen University, Guangzhou, Guangdong 510006, China

Received:2019-12-01 Online:2020-06-25 Published:2020-08-07
About author:CHEN Haowei received the B.S. degree in computer science from the School of Data and Computer Science, Sun Yat-sen University (SYSU), China in 2020. He is working towards the master’s degree in the School of Data and Computer Science, SYSU. His research interests include mobile deep computing, edge intelligence and deep learning.|ZENG Liekang received the B.S. degree in computer science from the School of Data and Computer Science, Sun Yat-sen University, China in 2018. He is currently pursuing the master’s degree with the School of Data and Computer Science, Sun Yat-sen University. His research interests include mobile edge computing, deep learning, and distributed computing.|YU Shuai received the Ph.D. degree from Pierre and Marie Curie University (now Sorbonne Université), France, in 2018, the M.S. degree from Beijing University of Post and Telecommunications, China, in 2014, and the B.S. degree from Nanjing University of Post and Telecommunications, China, in 2009. He is now a post-doctoral Research Fellow at the School of Data and Computer Science, Sun Yat-sen University. His research interests include wireless communications, mobile computing and machine learning.|CHEN Xu (chenxu35@mail.sysu.edu.cn) is a full professor in Sun Yat-sen University, China, and the vice director of National and Local Joint Engineering Laboratory of Digital Home Interactive Applications. He received the Ph.D. degree in information engineering from The Chinese University of Hong Kong in 2012, and worked as a post-doctoral research associate at Arizona State University, USA from 2012 to 2014, and a Humboldt Scholar Fellow at Institute of Computer Science of University of Goettingen, Germany from 2014 to 2016. He is currently an area editor of IEEE Open Journal of the Communications Society, an associate editor of the IEEE Transactions Wireless Communications, IEEE Internet of Things Journal and IEEE Journal on Selected Areas in Communications (JSAC) Series on Network Softwarization and Enablers.
Supported by:
the National Science Foundation of China(61972432);the Program for Guangdong Introducing Innovative and Entrepreneurial Teams(2017ZT07X355)

Abstract

Abstract:

Edge computation offloading allows mobile end devices to execute compute-intensive tasks on edge servers. End devices can decide whether the tasks are offloaded to edge servers, cloud servers or executed locally according to current network condition and devices’ profiles in an online manner. In this paper, we propose an edge computation offloading framework based on deep imitation learning (DIL) and knowledge distillation (KD), which assists end devices to quickly make fine-grained decisions to optimize the delay of computation tasks online. We formalize a computation offloading problem into a multi-label classification problem. Training samples for our DIL model are generated in an offline manner. After the model is trained, we leverage KD to obtain a lightweight DIL model, by which we further reduce the model’s inference delay. Numerical experiment shows that the offloading decisions made by our model not only outperform those made by other related policies in latency metric, but also have the shortest inference delay among all policies.

Key words: mobile edge computation offloading, deep imitation learning, knowledge distillation

CHEN Haowei, ZENG Liekang, YU Shuai, CHEN Xu. Knowledge Distillation for Mobile Edge Computation Offloading[J]. ZTE Communications, 2020, 18(2): 40-48.

Figures/Tables 7

Figure 1 Subtasks are offloaded to end device, edge server and cloud server respectively.

Figure 2 Deep imitation learning model for edge computing offloading. Given the offloading requirement S=(E, D, p1, b1, p2, b2) as the input, the deep imitation learning model can output the offloading decision I* = (a1, a2, a3, a4, a5, a6).

Figure 3 Compress model by knowledge distillation to get a lightweight model deployed to edge server.

Figure 4 Complete flowchart of our edge offloading framework based on DIL and KD.

Figure 5 Normalized end-to-end latency of offloading decisions made by our DIL model and baselines.

Figure 6 Normalized KD-DIL model and baselines when using a small training set.

Table 1 Inference delay of all models

Model Name	KD-DIL	Large DIL	DRL	Greedy	Optimal
Normalized Delay	0.17	0.51	119.72	1.00	122.54

References 16

1	YAP K H, CHEN T, LI Z, et al. A comparative study of mobile⁃based landmark recognition techniques [J]. IEEE intelligent systems, 2010, 25(1): 48–57. DOI:10.1109/mis.2010.12 DOI
2	JIANG J C, ANANTHANARAYANAN G, BODIK P, et al. Chameleon: scalable adaptation of video analytics [C]//2018 Conference of the ACM Special Interest Group on Data Communication. Budapest, Hungary, 2018:253–266. DOI:10.1145/3230543.3230574 DOI
3	Multi⁃access edge computing⁃standards for MEC [EB/OL].(2019⁃11⁃04)[2020⁃01⁃05].
	⁃clusters/technologies/multi⁃access⁃edge⁃computing
4	YU S, CHEN X, YANG L, et al. Intelligent edge: leveraging deep imitation learning for mobile edge computation offloading [J]. IEEE wireless communications, 2020, 27(1): 92–99. DOI:10.1109/mwc.001.1900232 DOI
5	ZHANG T H, MCCARTHY Z, JOW O, et al. Deep imitation learning for complex manipulation tasks from virtual reality teleoperation [C]//2018 IEEE International Conference on Robotics and Automation (ICRA). Brisbane, Australia, 2018: 1–8. DOI:10.1109/icra.2018.8461249 DOI
6	CHEN X, JIAO L, LI W Z, et al. Efficient multi⁃user computation offloading for mobile⁃edge cloud computing [J]. ACM transactions on networking, 2016, 24(5): 2795–2808. DOI:10.1109/tnet.2015.2487344 DOI
7	HE Y, ZHAO N, YIN H X. Integrated networking, caching, and computing for connected vehicles: a deep reinforcement learning approach [J]. IEEE transactions on vehicular technology, 2018, 67(1): 44–55. DOI:10.1109/tvt.2017.2760281 DOI
8	BA L J, CARUANA R. Do deep nets really need to be deep? [EB/OL].(2014⁃10⁃11) [2020⁃01⁃05].
9	HINTON G, VINYALS O, DEAN J. Distilling the knowledge in a neural network [EB/OL].(2015⁃03⁃09)[2020⁃01⁃10].
10	RAN X, CHEN H, ZHU X, et al. Deep decision: A mobile deep learning framework for edge video analytics [C]//IEEE INFOCOM 2018⁃IEEE Conference on Computer Communications. IEEE, 2018: 1421–1429
11	KANG Y P, HAUSWALD J, GAO C, et al. Neurosurgeon: collaborative intelligence between the cloud and mobile edge [J]. ACM SIGARCH computer architecture news, 2017, 45(1): 615–629. DOI:10.1145/3093337.3037698 DOI
12	CUERVO E, BALASUBRAMANIAN A, D⁃KCHO, et al. MAUI: Making smartphones last longer with code offload [C]//Proceedings of the 8th International Conference on Mobile Systems, Applications, and Services. San Francisco, USA, 2010: 49–62
13	TSOUMAKAS G, KATAKIS I. Multi⁃label classification [J]. International journal of data warehousing and mining, 2007, 3(3): 1–13. DOI:10.4018/jdwm.2007070101 DOI
14	YOU C S, ZENG Y, ZHANG R, et al. Asynchronous mobile⁃edge computation offloading: energy⁃efficient resource management [J]. IEEE transactions on wireless communications, 2018, 17(11): 7590–7605. DOI:10.1109/twc.2018.2868710 DOI
15	ZHOU Z, CHEN X, LI E, et al. Edge intelligence: paving the last mile of artificial intelligence with edge computing [EB/OL]. (2019⁃05⁃24)[2020⁃01⁃05]. DOI:10.1109/JPROC.2019.2918951 DOI
16	CHEN X, PU L, GAO L, et al. Exploiting massive D2D collaboration for energy⁃efficient mobile edge computing [J]. IEEE wireless communications, 2017, 24(4): 64–71. DOI:10.1109/MWC.2017.1600321 DOI

Knowledge Distillation for Mobile Edge Computation Offloading

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 7

References 16

Related Articles 0

Recommended Articles

Metrics