Improving Application Launch Performance in Smartphones Using Recurrent Neural Network

A. L. N. Martins, C. A. Duarte, J. Jeong. ICMLT 2018

[ACM]

Mobile phones became indispensable tools in our lives, with Android being the most used mobile OS. These devices depend on managing application lifecycles to improve launch performance, but the management of application processes is not done in an efficient way. The standard low memory killer, that is responsible for freeing memory, does not consider any user information and it frequently kills applications that are going to be launched. Context-awareness presents several possibilities to make mobile systems more efficient and user-driven in terms of user experience. In this paper, we introduce a context-based launcher using recurrent neural network (RNN), a special branch of neural networks capable of remembering dependencies, taking in consideration not just previous inputs, but also previous outputs, providing high accuracy without any extra sensor context. Our system guarantees that most of the applications are ready to use in the background, substantially improving the launch time enabling a better user experience. Experimental results demonstrate that the novel scheme can reduce application launch latency in a significant manner.