Machine Learning on Android Demystified
summary
Compares cloud vs on‑device ML and shows a runnable Android path: Kotlin examples using TensorFlow Lite with GPU delegate and NNAPI to run a pose‑estimation postnet model, image preprocessing, interpreter inputs/outputs and drawing keypoint coordinates with confidence scores; covers converting and optimizing models (including 8‑bit quantization to shrink size and cut latency), and using Firebase ML Kit/AutoML Vision Edge to host, update and switch between local and cloud models while weighing accuracy, latency, battery and privacy tradeoffs.
AI-written summary · about
chapters (17)
- 0:00Welcome and introduction
- 2:29What is machine learning
- 6:02ML use cases and pitfalls
- 9:18Cloud vs on-device ML
- 11:06On-device ML advantages
- 13:35ML model challenges
- 14:20TensorFlow Lite overview
- 17:49Ready-to-use TF Lite models
- 19:15TF Lite implementation steps
- 22:23Converting models to TF Lite
- 23:26Optimizing models for mobile
- 26:34Firebase ML Kit introduction
- 27:39ML Kit features overview
- 28:43Smart Reply with ML Kit
- 30:49AutoML Vision Edge
- 32:59Custom models with ML Kit
- 37:13Summary and key takeaways
related talks
47mMachine Learning on Android Demystified
47mMyths and Reality of Android Machine Learning
15mSimplified Machine Learning on mobile, browsers, and edge devices with on-device ML
39mAI/ML on mobile/android with TensorFlow
43mWhat’s new in Android machine learning
36mMachine Learning for Mobile Developers
09mMachine Learning on the Edge
30mMachine Learning in Android Apps with ML Kit