Smartphones could get smarter by learning user’s habits

Smartphone users could squeeze even more battery life out of their devices thanks to a new research methodology from the University of Essex.

Researchers at the School of Computer Science and Electronic Engineering have developed the first machine learning algorithm to optimize a mobile phone's performance and thermal behaviour based on the user's interaction with the phone.

This is the first time this methodology based on user behaviour has been proposed and in lab tests it outperformed existing methodologies in the top smartphones used for the purpose of performance and thermal behaviour.

Developed by a team led by computer scientist Somdip Dey, the methodology would mean the user would be able to use their phone for longer and improve the ongoing issues with smartphone battery life. The research paper was presented at a top electronic design conference, DATE 2020, and has already gained interest from other researchers keen to pursue the methodology in further development of resource optimization techniques in mobile phones.

“This is ground-breaking work and the first to propose a reinforcement learning based machine learning approach to optimize performance, energy consumption, and thermal behaviour in a mobile device by taking user’s behaviour with the device into consideration,“ explained Somdip, who pursued the work while working at the Samsung R&D Institute last year.

“By learning from the user’s behaviour, we have shown how smartphones could be developed to get even more battery life for the same usage.”

The team has developed a reinforcement learning based agent which tracks how an app is being used throughout the day. For example, a user might quickly scroll through the BBC News app while at work to check the headlines, which will require a higher FPS (frames per second) than when they spend more time on the app in the evening, slowly scrolling down and reading more stories in full.

The methodology means the agent realises the change in FPS for the app being used and tries to find the best operating frequency of CPU and GPU processors to cater for the change in app use behaviour whilst consuming the least amount of power and temperature gain in the device, which is a critical issue in mobile phones.