Machine Learning Basics

7073f145300a2755d821629427e0df51?s=47 Tom Walder
October 01, 2017

Machine Learning Basics

Machine Learning (ML) isn’t Skynet, but it *is* a type of Artificial Intelligence.

It’s certainly more easily accessible than ever - and could add great value to your software.

We’ll cover the basic principles of the Machine Learning process: DATA > LEARNING > PREDICTION

There are easily accessible, pre-trained machine learning REST APIs for image, text, video and voice analysis. These can be a real short-cut to taking advantage of ML quickly in your applications.

We’ll look at some of these APIs and their application in real-world software.

When your problem is harder, more niche or needs some customisation, you’ll need to “train your own model”.

We’ll discuss the importance of data in Machine learning - the starting point for any new model.

There are some great open-source tools (in PHP as well as the popular TensorFlow in Python) for building and training your own models - and with scalable, low-cost cloud servers you can train new models quickly in the cloud.

7073f145300a2755d821629427e0df51?s=128

Tom Walder

October 01, 2017
Tweet

Transcript

  1. 4.

    Artificial Intelligence Intelligence exhibited by machines, rather than humans or

    other animals Wikipedia Machine Learning Giving computers the ability to learn without being explicitly programmed Arthur Samuel (via Wikipedia)
  2. 5.

    Artificial Intelligence Intelligence exhibited by machines, rather than humans or

    other animals Wikipedia Machine Learning Giving computers the ability to learn without being explicitly programmed Arthur Samuel (via Wikipedia)
  3. 9.

    Applications of ML • Google search • Facial recognition •

    Apps (translation) • Self-driving cars
  4. 10.

    Applications of ML • Google search • Facial recognition •

    Apps (translation) • Self-driving cars • Terminators?
  5. 11.

    A Learning Computer? How do you teach a computer? How

    can it learn and improve? MATHS + DATA Software! ______
  6. 12.

    Types of ML Algorithm Supervised Know outcomes in data •

    Classification Diseased/Healthy Stop/Go • Regression Height: 186cm Share price: 50p Unsupervised Discovery • Clustering Discover inherent groups • Association People who buy X also buy Y
  7. 13.

    ML Lifecycle Overview Data Learning Prediction Choose & train a

    mathematical model. Iterative, guided process. Source & prepare your data Executing your model on new, unseen data.
  8. 14.

    Data Data Learning Prediction Identify features that are important to

    your problem Data format - text, images, video, audio Source, understand, clean, filter, normalise Split into training & evaluation data The quality of your final model will depend on your data.
  9. 17.

    Head (mm) Temp (C) Good pint? 14 3.5 Yes 18

    14 No 28 5 No 16 4 Yes 13 5 Yes 19 8.2 No 50 0 Not sure Data
  10. 18.

    Learning Data Learning Prediction Select & train a mathematical model

    Iterative, guided process Computationally intensive Human input is key Cloud, training at scale, GPUs, TPUs Output is a set of algorithms, weights Iterative performance evaluation.
  11. 20.
  12. 23.

    Most Popular ML Library • Python • Originally Google •

    Open Source in 2015 • Bas become the most widely adopted ML lib • Large range of tools, visualisations • Works seamlessly with cloud- based training and serving products from Google • Runs on devices
  13. 25.

    Google ML REST APIs • Natural Language • Speech •

    Translation • Vision • Video Intelligence Pre-trained machine learning REST APIs for image, text, video and voice analysis.
  14. 26.

    The End • Machine Learning Basics • Tom Walder /

    @tom_walder • Feedback / https://joind.in/talk/58f6b • Resources • https://github.com/php-ai/php-ml • https://www.tensorflow.org/ • https://cloud.google.com/products/machine-learning/ • https://cloud.google.com/dataprep/ Bye