On-device machine learning in mobile and web apps
summary
TensorFlow Lite launches in 2017 with a converter that turns cloud models into efficient on‑device models and delivers millisecond inference latency for real‑time video and audio. Turnkey APIs via ML Kit cover common use cases while custom work uses the TF Lite runtime, task library and Model Maker. The Model Optimization Toolkit offers quantization reduces model size to about 25% and other reductions. Web use runs via TensorFlow.js and TFLite models, on‑device training prototypes target a Q3 Colab, and TensorFlow Lite Micro already ships in production (OK Google).
AI-written summary · about
chapters (10)
- 0:00Intro
- 1:10What is ondevice ML
- 5:00Use cases
- 8:35ML in web apps
- 10:10Model protection
- 18:23Ondevice training
- 19:29MLKit vs TensorFlow Lite
- 23:14Edge computing and ML
- 25:17Models getting stolen
- 26:44Reducing model size
related talks
34mTensorFlow Lite
32mAI for Mobile and IoT Devices: TensorFlow Lite
39mMachine Learning on Your Device: The Options
17mDeploy a custom Machine Learning model to mobile
46mTensorFlow Lite in Android
47mMachine Learning on Android Demystified
29mFrom TensorFlow to ML Kit: power your Android application with ML
39mFrom TensorFlow to ML Kit: power your mobile application with machine learning