Don't let your app drain your users' battery
summary
Adaptive Battery uses app standby buckets (active, working set, frequent, rare) and an on-device ML model trained on millions of app-open sequences to predict next-open probability; apps in rare get throttled for jobs, alarms, FCM high-priority caps and networking, with restrictions lifted while charging. The model is a two-layer convolutional net plus feedforward running via on-device TensorFlow. Devices also enforce background limits for apps not targeting Oreo and for apps holding wakelock >1 hour. The UI exposes battery saver, an Adaptive Battery API and adb testing commands.
AI-written summary · about
chapters (14)
- 0:00Intro
- 2:49Efforts to improve battery
- 5:21Challenges Remain
- 7:09How to address them?
- 8:49App Standby Buckets
- 11:32Machine Learning
- 13:29The Model
- 14:18Fairness and Privacy
- 15:04Adaptive Battery
- 16:08Battery Saver
- 17:57Battery Settings
- 19:29Background Restricted Criteria
- 20:13Background Restrictions
- 23:08How do I do background work?
related talks
21mImproving battery life with restrictions
33mIs my application taking up too much power?
34mPowering up your app performance & stability using App Vitals
30mApp Standby vs. Doze: Understanding System Sleep
28mBest practices for running background work on Android
35mHow to survive on Android and don't get killed
22mPower optimization for Android
48mHow to Build Performance-Driven Android Apps