Simplified Machine Learning on mobile, browsers, and edge devices with on-device ML
summary
Demonstrates true on-device training with a shipped Android demo that collects images and trains a classifier entirely on phone using TensorFlow Lite and variables, signatures, gradients and optimizers, with a practical example collecting 10 photos for each of four classes after exploring a 40‑category dataset on-device training demo with 40-category dataset Highlights runtime and deployment changes including TensorFlow Lite runtime in Google Play Services, browser support for quantized models, a 30% ARM64 speedup, 5x x86 claim from XNNPACK, tools for selective pruning and quantization debugging, a GPU compatibility checker and analyzer, and a new MoveNet 17-point pose model and reference apps.
AI-written summary · about
chapters (8)
- 0:00Start
- 1:47ML Kit - barcode scanning
- 2:00ML Kit - pose detection
- 2:16ML Kit-language identification
- 5:49Example of on-device training flow
- 8:14Updates to TensorFlow Lite in Google Play Services
- 13:02Target Aware Authoring (experimental feature) Discover TF TensorFlow Lite conversion issues at model authoring time • Verify if your model is compatible with the GPU delegate at authoring time
- 13:55Pose Estimation - MoveNet model
related talks
39mMachine Learning on Android Demystified
39mAI/ML on mobile/android with TensorFlow
46mFire Up Your Smartphone Neurons with On-Device Deep Learning
19mOn-device machine learning: TensorFlow on Android
47mMachine Learning on Android Demystified
42mCreating Smarter Applications
39mMachine Learning on Your Device: The Options
47mMyths and Reality of Android Machine Learning