WWDC 2017

Table of Contents

Machine Learning (CoreML) - Tuesday
- Why ?
- Training
- ML Frameworks
- Run on device
- Model
- Development

Machine Learning (CoreML) - Tuesday

Session video and resources: https://developer.apple.com/videos/play/wwdc2017/703/

Why ?

Real time image recognition
Text prediction
Entity recognition
Style transfer
Speaker identification and many more...

Training

Learning algorithm -> Model
Input -> Model -> Desired Output
Challenges
- Correctness
- Performance
- Energy efficiency

ML Frameworks

VisionKit
- Object tracking
- Face detection
NLP (Natural language processing)
- Language identification
- Named entity recognition API
Core ML
- Music tagging (tag parts of the music with data)
- Image captioning (image -> text)
All these are powered with Accelerate and MPS
- High performance math

Run on device

- Privacy
- Prevent data cost
- Server cost
- Always available (24/7)

Model

Function learned from data
Observed inputs
Predicts outputs
Single document
Public format
Sample models
- https://developer.apple.com/machine-learning
- Core ML models
- Ready to use
Convert to Core ML using python packages (This is fully open source)

Development

Add your ML Model to your project -> autogeneration to Swift file