Work

Local Distributed Mobile Computing System for Deep Neural Networks

Year: 2017
Type: article
Abstract: Nowadays, Deep Neural Networks (DNN) are emerging as an excellent candidate in many ap- plications (e.g., image classification, object detection and natural language processing). Though ubiquitously utilized in many fields, DNN models are generally hard to be deployed on resource-constrained devices (e.g., mobile devices). In the prior arts, the research topics mainly focus on client-server computing paradigm or DNN model compression, which, respectively, ask for either outside infrastructure support or special iterative training phases. In this work, I propose a lo- cal distributed mobile computing system for the testing phase of DNNs called MDNN, short for Mobile Deep Neural Network. MDNN partitions already trained DNN models onto several mo- bile devices with the same local wireless network to accelerate DNN computations by alleviating device-level computing cost and memory usage. Two model partition schemes are also designed to minimize non-parallel data delivery time, including both wakeup time and transmission time. Experimental results show that when the number of worker nodes increases from 2 to 4, MDNN can accelerate the DNN computation by 2.17-4.28. Besides the parallel execution, the performance speedup also partially comes from the reduction of the data delivery time, e.g., 30.02% w.r.t. con- ventional 2D-grids partition. Furthermore, a model compression using group lasso is utilized for simultaneously alleviating computing cost and transmission cost.
Author: Mao, Jiachen
Language: English
Open Access status: green
OpenAlex ID: W6992571650
FWCI:
Cites: 15
Cited by: 2
Related to: 10