Mobiconf 2018: Vladimir Alekseichenko, DataWorkshop
summary
Argues that running models on-device often uses less energy than uploading data and preserves privacy, shows converted models (ResNet/Inception/ResNet50 example) running via Core ML, ML Kit and TensorFlow on phone and Raspberry Pi, notes typical model sizes ~100+ MB and reduction via 8‑bit/4‑bit quantization, cites Netflix saving ~$1B from recommendations and a 1% detection lift worth ~$2B, explains batching for speed and mentions Google’s federated learning idea and ML Kit + Firebase integration.
AI-written summary · about
chapters (9)
- 0:00Introduction to the fourth industrial revolution
- 3:07Real-world AI applications today
- 7:06Machine learning fundamentals explained
- 11:31Key components of machine learning systems
- 17:17Frameworks and tools for ML development
- 21:18Running ML models on mobile devices
- 27:24Optimizing models for mobile deployment
- 33:15Federated learning on mobile networks
- 38:19Ethical and practical considerations
related talks
29mPushing Machine Learning into your app When TensorFlow and MLKit rule the world
47mMachine Learning on Android Demystified
21mGetting the best of on-device ML with ML Kit
42mCreating Smarter Applications
29mFrom TensorFlow to ML Kit: power your Android application with ML
21mThings I learned running neural networks on microcontrollers
39mFrom TensorFlow to ML Kit: power your mobile application with machine learning
54mSmarter Everything with ML & the IoT