Energy-efficient prediction of smartphone unlocking

Abstract

We investigate the predictability of the next unlock event on smartphones, using machine learning and smartphone contextual data. In a two-week field study with 27 participants, we demonstrate that it is possible to predict when the next unlock event will occur. Additionally, we show how our approach can improve accuracy and energy efficiency by solely relying on software-related contextual data. Based on our findings, smartphone applications and operating systems can improve their energy efficiency by utilising short-term predictions to minimise unnecessary executions, or launch computation-intensive tasks, such as OS updates, in the locked state. For instance, by inferring the next unlock event, smartphones can pre-emptively collect sensor data or prepare timely content to improve the user experience during the subsequent phone usage session.

“It's Like Being Gone For A Second”

“…and that interactions with the lock screen lasted around 13.5 seconds [5]. The general uniformity of these patterns is also put into evidence by experimental software that can predict the next unlocking of the smartphone with reasonable accuracy [51].…”

“…This may be because there was no relevant information available to be displayed to them, or because the options displayed were inappropriate for the situation they are in (watching a video with sound may not be desirable when queuing at the supermarket checkout, replying to an Email may take too long). If existing approaches like [51] can be further developed to efficiently predict that users are waiting for something external when unlocking the phone, these would be excellent instances to apply brief ESM or Slide to X treatments, not only increasing acceptance because they would not constitute disruptions in these cases, but also by leveraging the user's desire to reduce the time spent 'idling' and putting it to efficient use instead.…”

An Intelligent Optimization Strategy Based on Deep Reinforcement Learning for Step Counting

“…Lou et al [11] combined the intermediate threshold method to determine the wave peak value. Lou et al [12] proposed an adaptive peak detection method, and according to the set threshold, the normal state and abnormal state are determined, and then, different neighborhood windows are set for different states, and the wave peaks are counted in the window. Liu and Yang [13] designed an adaptive time window to determine the wave crest and trough simultaneously through the adaptive double threshold.…”

“…, T. do (10) G t � 􏽐 T k�t r k . (11) According to formula (4), optimizing parameters (12) Updating parameters: θ←θ − α • ∇L(θ). (13) end for ( 14) end for ALGORITHM 1: e proposed approach.…”

Energy-aware JPEG image compression: A multi-objective approach

“…By making a large-scale data set, they made the Energy Emulation Toolkit (EET), which lets developers compare the energy needs of their Apps to real user energy traces. In another work, [15] used machine learning (ML) and smartphone environment data to determine if a smartphone's next unlock event can be predicted. They showed that it is possible to predict when the next unlock event will happen by doing a 2-week field study with 27 people, leading to improve accuracy and saving energy by using only software-related background data.…”