TensorFlow Neural Networks on iOS
summary
A concrete walkthrough of training, shrinking and shipping image models: TensorFlow training on a growing specimen dataset yields 87% accuracy and a mobile classifier that serves identifications in about 200 ms latency. Practical notes cover data balancing, augmentation, using MobileNet or SqueezeNet instead of inception v3 to cut size (example MobileNet ~14 MB, TensorFlow adds ~40 MB), freezing/optimizing .pb files, embedding models in iOS and delivering over‑the‑air updates via Core ML compile.
AI-written summary · about
chapters (8)
- 0:00Intro
- 1:14Using TensorFlow, CoreML, Metal Performance Shaders, Accelerate BNNs, Keras and What the Heck is a Neural Network Anyway?
- 2:34Fieldguide
- 11:53What is a Neural Network Anyway?
- 17:27Network Training Basics
- 28:18Neural Networks on iOS
- 37:32Always runs on GPU
- 38:48Delivering Updates
related talks
27mMachine Learning and Mobile
43mTensorFlow for Mobile Developers
30mTensorFlow for Mobile Heroes
40mApplied TensorFlow in Android apps
25mGetting Started with Machine Learning in your iOS apps
42mWhat's New with TensorFlow?
29mRunning Deep Learning Models on Mobile with optimised speed
46mTensorFlow for Mobile Poets