学术报告:BottleNet++: Effective Feature Compression and Transmission for Edge Inference
发布时间: 2019-11-20   浏览次数: 167


Title:

BottleNet++: Effective Feature Compression and Transmission for Edge Inference


Jun Zhang

Assistant Professor

Department of Electronic and Information Engineering at the Hong Kong Polytechnic University (PolyU)


时间20191120日 下午330-530

地点南京邮电大学物联网研究院会议室


Abstract

 Deep learning has achieved great successes in many application domains, such as computer vision, image processing, and natural language processing. However, deploying powerful deep learning models on resource-constrained mobile devices (e.g., wearable or IoT devices) faces great challenges. Recently, edge AI techniques that rely on the emerging mobile edge computing platforms have been proposed, which enable effective training and inference on mobile devices. This talk will introduce recent advancements on device-edge co-inference, which splits a deep neural network (DNN) at a mobile device and an edge computing server. An end-to-end architecture, named BottleNet++, will be introduced, which jointly considers neural network splitting, intermediate feature compression and transmission. By exploiting the strong sparsity and the fault-tolerant property of the intermediate feature in a DNN, BottleNet++ achieves a much higher compression ratio than existing methods. This in turn enables splitting a deep neural network at earlier layers, which leads to up to 3x reduction in on-device computation compared with other compression methods. An interesting finding is that analog communication may enjoy advantages in edge AI systems.

Biography

 Dr. Jun Zhang received the Ph.D. degree in Electrical and Computer Engineering from the University of Texas at Austin. He is currently an Assistant Professor in the Department of Electronic and Information Engineering at the Hong Kong Polytechnic University (PolyU). His research interests include wireless communications and networking, mobile edge computing and edge learning, distributed learning and optimization, and big data analytics.

 Dr. Zhang co-authored the books Fundamentals of LTE (Prentice-Hall, 2010), and Stochastic Geometry Analysis of Multi-Antenna Wireless Networks (Springer, 2019). He is a co-recipient of the 2019 IEEE Communications Society & Information Theory Society Joint Paper Award, the 2016 Marconi Prize Paper Award in Wireless Communications (the best paper award of IEEE Transactions on Wireless Communications), and the 2014 Best Paper Award for the EURASIP Journal on Advances in Signal Processing. Two papers he co-authored received the Young Author Best Paper Award of the IEEE Signal Processing Society in 2016 and 2018, respectively. He also received the 2016 IEEE ComSoc Asia-Pacific Best Young Researcher Award. He is an Editor of IEEE Transactions on Wireless Communications, IEEE Transactions on Communications, and Journal of Communications and Information Networks.


                      南京邮电大学通信与信息工程学院

      物联网研究院  通信技术研究所

2019.11.18