Abstract: Deep Neural Network (DNN) has been a great success in many areas recently. It has achieved an advanced performance in applications such as image classification, speech recognition, and time series forecasting. However, when the data is getting bigger, the performance computation on a single CPU becomes worse.  One approach to tackle this challenge is to distribute the DNN workload into several machines. This research uses mobile cloud computing as a joint distributed environment to run the DNN application. An image recognition problem used as a test case will be conducted on both mobile and cloud platforms. Latency time and energy consumption measured in each DNN layer and used as the parameters to allocating efficiently the computational task of each layer in to suitable cores of mobile devices or cloud in Mobile Cloud Computing (MCC) environment. The joint platform can achieve improvements up to 73% in latency and 56% in energy consumption. As an addition, we apply lossless compression to reduce the influence of the communication between layer.

Keywords:  Deep Neural Network, AlexNet, Image Recognition, Mobile Cloud Computing, Task Scheduling.

PDF | DOI: 10.17148/IJARCCE.2020.9902

Open chat
Chat with IJARCCE