AI is ready for you. Are you ready for AI?
summary
A practical guide to get ML running on Android with code on GitHub and a working image classifier using Inception, MobileNet and custom food nets, showing that node count, not just FLOPs, drives latency and that MobileNet is much smaller and faster in practice. Steps include freezing graphs, converting to TensorFlow Lite, stripping unsupported ops because Java client cannot train, matching preprocessing and color spaces exactly, using quantized weights to cut model size, optionally downloading models at runtime, and targeting Android 8.1+ to leverage NNAPI 4x speedups.
AI-written summary · about
chapters (9)
- 0:00AI on mobile intro
- 1:51AI in daily life
- 4:28ML development challenges
- 6:14TensorFlow Lite production
- 9:12Simple ML on Android
- 12:43Optimizing models for mobile
- 16:48Model structures vs performance
- 19:23Graph transformation tools
- 21:55Summary
related talks
39mAI/ML on mobile/android with TensorFlow
39mMachine Learning on Android Demystified
47mMyths and Reality of Android Machine Learning
38mNEURAL NETWORKS ON YOUR MOBILE PHONE. DIVING INTO THE NN API
40mApplied TensorFlow in Android apps
40mWho’s afraid of Machine Learning? First steps with TensorFlow for Android.
47mMachine Learning on Android Demystified
32mAI for Mobile and IoT Devices: TensorFlow Lite