Upgrade to Pro — share decks privately, control downloads, hide ads and more …

From Zero to ML on Google Cloud Platform

From Zero to ML on Google Cloud Platform

A quick start into your machine learning journey on Google Cloud Platform

Olayinka Peter Oluwafemi

December 01, 2018
Tweet

More Decks by Olayinka Peter Oluwafemi

Other Decks in Technology

Transcript

  1. From Zero to ML on Google Cloud Platform +OlayinkaPeterOluwafemi @olayinkapeter_

    Organizer, Google Cloud Developer Community, Ado-Ekiti, Nigeria GDG DevFest Makurdi 2018
  2. Generic task Custom task Stuff someone has solved before Very

    specific to your dataset Input to prediction
  3. Generic task Custom task Stuff someone has solved before Very

    specific to your dataset “puppy” “Sylvester” Input to prediction
  4. Generic task Custom task Stuff someone has solved before Very

    specific to your dataset Pre-trained models (Machine Learning APIs) Cloud Machine Learning Engine Application developers Data scientists and ML practitioners
  5. Generic task Custom task Stuff someone has solved before Very

    specific to your dataset Pre-trained models (Machine Learning APIs) Cloud Machine Learning Engine Application developers Data scientists and ML practitioners Cloud AutoML Lets you train your own custom machine learning models without writing model code
  6. • Pre-trained models can accomplish the following ML tasks Cloud

    Translation Cloud Video Intelligence Cloud Speech to Text & Text to Speech Cloud Vision Cloud Natural Language
  7. Powerful image analysis Google Cloud Vision API allows you to

    explore powerful image analysis with • Pre-trained models • Ability to build custom models using AutoML Vision Easy-to-use REST API Cloud Vision
  8. Sometime in late 2017, I published Toodoo, a Firebase-powered app

    that basically lets you add tasks you want to be reminded of, but with some Machine Learning capabilities.
  9. Toodoo is not special. Except that it solves the problem

    of: Users spending large amount of time and/or having trouble in trying to read small printed material in a voluminous page
  10. Hence, allowing users to: • Hunt for text material (in

    languages such as Chinese, English, Finnish, French, German, Japanese, Korean, Portuguese & Spanish) with a device camera to capture an image. • Process the image and extract the text from it using powerful Google Cloud vision Optical Character Recognition (OCR) support. • Copy the extracted image to clipboard, and add them as to-dos
  11. Easily detect broad sets of objects in your images With

    AutoML Vision, you can create custom models that highlight specific concepts from your images. This enables use cases ranging from categorizing product images to diagnosing diseases. Cloud Vision API also lets you
  12. Extracts and identifies text Optical Character Recognition (OCR) enables you

    to detect text within your images or via the camera, along with automatic language identification. Remember Toodoo?
  13. Power of the web Vision API uses the power of

    Google Image to find topical entities like celebrities, logos, or news events. Millions of entities are supported, so you can be confident that the latest relevant images are available.
  14. Moderates content With the power of Google SafeSearch, one can

    easily moderate content and detect inappropriate content from crowd-sourced images. That is, detect different types of inappropriate content, from adult to violent content.
  15. Google Cloud pre-trained ML APIs allows you to do all

    that machine learning, but there’s more… What if there’s need to train these APIs on your own custom data?
  16. • Cloud AutoML allows you to train custom machine learning

    models for vision, natural language, and translation without writing model code Cloud Translation Cloud Vision Cloud Natural Language
  17. Identify your own puppy among hundreds of puppies “Sylvester” “Driver”

    here could be referred to as ‘taxi driver’, for example Importance of specific prediction or translation
  18. So, I love Wilson’s Lemonade, and tried out a very

    small amount of photos of it’s three different types.
  19. Note that each label should have at least 100 images

    for best results. However, for this sample, I used about 10 each (do not try this at home).
  20. Tested the custom model anyway :D Here’s a wrong prediction

    for two apparent reasons: 1. Not enough data 2. The Tea label had a number of false negatives We can, of course, remove the confusing images, add more data, and train again. The Pink label had a better precision-recall tradeoff curve, hence a new image was predicted right.
  21. AutoML is great, yeah! However, what if you have a

    custom task that requires more control over your model type and its inputs?
  22. • Cloud ML Engine allows you to build, train, and

    serve custom models with your own data
  23. • Use pre-trained APIs to accomplish common ML tasks like

    image analysis, natural language processing, or translation • Use AutoML to train a model on your own data without writing model code • Use ML Engine to build a model with your data for custom tasks Three things to remember