Domo Arigato, Mr. Roboto: Machine Learning with Ruby

Slides for my RubyConf 2016 talk on machine learning.

Transcript

1. Dōmo arigatō, Mr. Roboto:
   Machine Learning with Ruby
   # Eric Weinstein
   # RubyConf 2016
   # Cincinnati, Ohio
   # 10 November 2016
   

   View Slide

2. for Joshua
   

   View Slide

3. Part 0: Hello!
   

   View Slide

4. About Me
   eric_weinstein = {
   employer: 'Hulu',
   github: 'ericqweinstein',
   twitter: 'ericqweinstein',
   website: 'ericweinste.in'
   }
   30% off with RUBYCONF30!
   

   View Slide

5. Agenda
   • What is machine learning?
   • What is supervised learning?
   • What’s a neural network?
   • Machine learning with Ruby and the MNIST
   dataset
   

   View Slide

6. Part 1: Machine Learning
   

   View Slide

7. View Slide

8. What’s machine learning?
   

   View Slide

9. In a word:
   

   View Slide

10. Generalization
    

    View Slide

11. What’s Supervised Learning?
    Classification or regression, generalizing from
    labeled data to unlabeled data
    

    View Slide

12. Features && Labels
    • Raw pixel features (vectors of intensities)
    • Digit (0..9)
    

    View Slide

13. Features && Labels
    • Raw pixel features (vectors of intensities)
    • Digit (0..9)
    

    View Slide

14. Image credit: https://www.tensorflow.org/versions/r0.9/tutorials/mnist/
    beginners/index.html
    

    View Slide

15. What’s a neural network?
    

    View Slide

16. Image credit: https://github.com/cdipaolo/goml/tree/master/perceptron
    

    View Slide

17. Image credit: https://en.wikipedia.org/wiki/Artificial_neural_network
    

    View Slide

18. Part 2: The MNIST Dataset
    

    View Slide

19. Our Data
    • Images of handwritten digits, size-normalized
    and centered
    • Training: 60,000 examples, test: 10,000
    • http://yann.lecun.com/exdb/mnist/
    

    View Slide

20. Image credit: https://www.researchgate.net/
    

    View Slide

21. How’d We Do?
    • Correct: 9328 / 10_000
    • Incorrect: 672 / 10_000
    • Overall: 93.28% accuracy
    

    View Slide

22. Developing the App
    

    View Slide

23. Front End
    submit() {
    fetch('/submit', {
    method: 'POST',
    body: this.state.canvas.toDataURL('image/png')
    }).then(response => {
    return response.json();
    }).then(j => {
    this.setState({
    prediction: j.prediction
    });
    });
    }
    

    View Slide

24. Front End
    render() {
    return(
    
    
    
    
    
    
    
    
    );
    }
    

    View Slide

25. Back End
    train = RubyFann::TrainData.new(inputs:
    features, desired_outputs: labels)
    fann = RubyFann::Standard.new(num_inputs: 576,
    hidden_neurons: [300], num_outputs: 10)
    fann.train_on_data(train, 1000, 10, 0.01)
    

    View Slide

26. STOP
    #demotime
    

    View Slide

27. Summary
    • Machine learning is generalization
    • Supervised learning is labeled data ->
    unlabeled data
    • Neural networks are awesome
    • You can do all this with Ruby!
    

    View Slide

28. Takeaways (TL;DPA)
    • We can do machine learning with Ruby
    • Contribute to tools like Ruby FANN
    (github.com/tangledpath/ruby-fann) and sciruby
    (http://sciruby.com/)
    • Check it out: http://ruby-mnist.herokuapp.com/
    • PRs welcome! github.com/ericqweinstein/ruby-
    mnist
    

    View Slide

29. Thank You!
    

    View Slide

30. Questions?
    eric_weinstein = {
    employer: 'Hulu',
    github: 'ericqweinstein',
    twitter: 'ericqweinstein',
    website: 'ericweinste.in'
    }
    30% off with RUBYCONF30!
    

    View Slide