What’s new in Android machine learning
summary
Android announces concrete on-device ML pieces: Android runs on 2.5 billion devices and exposes the Neural Networks API for GPU/DSP/NPU acceleration; measured wins include 9x faster Face Detection on a MediaTek Helio P90 and 4x energy savings on Snapdragon 855 for OCR. New APIs: Speech Recognizer (implemented via a seven-step listener flow), CameraX four-step flow for image analysis, ML Kit growth at 60% month-to-month with face detection 18x faster and 13–24% more accurate, AutoML Vision Edge to export custom TF Lite models, and a demo of a TF Lite 620-byte model successfully running Celsius→Fahrenheit inference on-device.
AI-written summary · about
chapters (30)
- 0:00Introduction
- 1:15What is machine learning
- 4:55Great Up School
- 6:15On Device
- 8:19Machine Learning Tools
- 9:49Most Common Questions
- 12:53Android Platform
- 13:17TexttoSpeech
- 14:47Where is this API
- 16:05Implementation
- 18:20Camera API
- 21:57Additional sessions
- 22:29Dawn Chen
- 22:55MLK
- 23:46MLKid
- 25:30Object Detection
- 26:10Mobile Translate
- 26:46Image Labelling
- 27:34Automail
- 28:04Market Automail
- 28:20Image labeling
- 28:52Mobile app classification
- 29:10Model training
- 29:50Flower classifier
- 31:30New mature design guidelines
- 34:26Neural Network
- 34:45Simple Neural Network
- 36:57Model Size
- 39:03Example
- 41:13Outro
related talks
13mWhat's new in Android machine learning
47mMyths and Reality of Android Machine Learning
10mWhat's new in Android Machine Learning
43mMyths and Reality of Android Machine Learning
39mMachine Learning on Android Demystified
36mMaking intelligent Android apps
15mSimplified Machine Learning on mobile, browsers, and edge devices with on-device ML
39mAI/ML on mobile/android with TensorFlow