Creating Smarter Applications
summary
Shows how to add real on‑device ML using TensorFlow Lite and Google ML Kit (Firebase) with cloud labels ~40,000 vs on‑device ~10,000, plus Colab and TensorBoard for training and monitoring. Demonstrates building a custom classifier by retraining Inception v3 with transfer learning on roughly 200 Dalek images (author notes 300–500 is typical), quantizing 32→8 bit weights (≈1–2% accuracy loss) and shipping a working offline Dalek detector that runs fast on the device.
AI-written summary · about
chapters (18)
- 0:00Introduction and speaker intro
- 1:55AI and machine learning history
- 3:45Traditional programming vs ML shift
- 5:35Why use machine learning
- 8:47MLKit and cloud solutions
- 11:47Benefits of offline ML
- 13:15Demo of offline object detection
- 16:51Building custom offline models
- 19:45TensorFlow for model creation
- 24:29CNN and transfer learning
- 27:23Model optimization for mobile
- 30:36Training and validation process
- 31:42Converting models with TF Lite
- 32:43Deploying to Android
- 35:14Live demo and implementation
- 36:21Key lessons and best practices
- 39:17Community and resources
- 40:42AutoML and future trends
related talks
29mFrom TensorFlow to ML Kit: power your Android application with ML
15mSimplified Machine Learning on mobile, browsers, and edge devices with on-device ML
29mPushing Machine Learning into your app When TensorFlow and MLKit rule the world
21mGetting the best of on-device ML with ML Kit
39mMachine Learning on Android Demystified
21mLive Coding: Adding Machine Learning to Android Things in 10 Mins
24mML Kit: Machine learning for mobile developers
47mMachine Learning on Android Demystified