O'Reilly logo
live online training icon Live Online training

AI on the Google Cloud Platform First Steps

An Introduction to TensorFlow on the Cloud

Topic: System Administration
Micheal Lanham

This course takes a hands-on introduction to building practical AI on the Google Cloud Platform (GCP). We will start with developing fundamental knowledge of deep learning. Taking an in-depth look at how the core of deep learning works by playing simple puzzle-like games. These games will teach you the inner workings of the perceptron, multi-layered perceptron, autoencoders, convolutional and recurrent networks. Many newcomers to TensorFlow and deep learning struggle to understand these core concepts. Each new section of the course teaches fundamental core concepts through the use of a simple puzzle game. Participants will be taught the deep learning theory behind the game and then encouraged to play the game/puzzle to completion.

From playing games, we will progress to working with Google Colaboratory (Colab). Colab is a web platform that allows developers to test and build AI applications on the cloud. We will cover the basics and various tips and tricks of working with Colab. Then we progress to building deep learning models with TensorFlow. Starting by building simple analysis models on tabular data. Moving up to understanding examples that perform image analysis with convolution. Following that with analyzing time series or text data with recurrent networks. Finishing the course with an introduction to sequence to sequence learning.

By the end of the course the attendees should be comfortable with core deep learning concepts. They should also be comfortable enough to start to build their own working models with TensorFlow on Colab.

What you'll learn-and how you can apply it

By the end of this live, hands-on, online course, you’ll understand:

  • The fundamentals of deep learning. Including optimization, backpropagation and activation functions in various settings.
  • The difference between supervised and unsupervised learning.
  • How deep learning architectures like multi-layered perceptron, autoencoders, convolutional and recurrent networks function internally.
  • How to build more complex architectures like sequence to sequence models.
  • What analysis methods are, regression or classification, and how and when to use them.
  • The need for training and test data and how to manage over and underfitting of models.
  • Methods to manage and manipulate data for inception into deep learning models. (Text and Images)
  • When and for what problems to use various deep learning architectures on.
  • The required libraries and dependencies for building AI with Python.

And you’ll be able to:

  • Run TensorFlow with Python code on Colab.
  • Develop and use deep learning models.
  • Build your own cutting edge AI applications.
  • Take your first step in becoming a machine learning developer.
  • Prepare to take the new TensorFlow Certification exam from Google. All the deep learning content in this course covers the basic material in the exam.

This training course is for you because...

  • You’re a developer.
  • You work with data in all forms.
  • You want to become a machine learning developer.

Prerequisites

  • A basic understanding of linear algebra, statistics, and calculus
  • A working knowledge of Python 3
  • Experience using Python to work with data using libraries like pandas and NumPy (useful but not required)

Recommended preparation:

Recommended follow-up:

About your instructor

  • Micheal Lanham is a proven software and tech innovator with 20 years of experience. During that time, he has developed a broad range of software applications in areas including games, graphics, web, desktop, engineering, artificial intelligence, GIS, and machine learning applications for a variety of industries as an R&D developer. At the turn of the millennium, Micheal began working with neural networks and evolutionary algorithms in game development. He has used those skills and experiences working as a GIS and BigData/enterprise architect to enhance and gamify a variety of engineering and business applications. Since late 2016 Micheal has been an avid author and presenter by giving his knowledge back to the community. Currently, he has completed 6 technical books, 2 on extended reality XR, 3 on deep reinforcement learning and AI and one on developing sound/music for games. He is known for many areas in AI and software development but currently specializes in developing deep learning generators and advanced AI with deep reinforcement learning. Micheal resides with his family in Calgary, Canada and is currently writing other books on AI, programming and software development.

Schedule

The timeframes are only estimates and may vary according to how the class is progressing

Introduction to AI and deep learning (55 minutes)

  • Presentation: Introduction to AI, deep learning, and the perceptron; supervised learning; the perceptron update equation; loss and backpropagation; unsupervised learning and autoencoders
  • Hands-on exercise: Play the first level of the deep learning game on the perceptron; play the first level again, this time with batching; play the second level on autoencoders
  • Q&A

Break (5 minutes)

Building AI on GCP (55 minutes)

  • Group discussion: What’s your experience with ML and TensorFlow?
  • Presentation: Introduction to Google Colab; building your first model to perform analysis on tabular data; building an autoencoder model against MNIST handwritten digits
  • Hands-on exercise: Explore the importance of tuning hyperparameters on a basic model; determine what can affect model training performance; tune hyperparameters of the model and explore how autoencoders work firsthand
  • Q&A

Break (5 minutes)

Understanding image analysis with convolution (55 minutes)

  • Presentation: Introduction to convolutional networks; how they work; upgrading the autoencoder example from last session to use the MNIST; fashioning your dataset; convolutional layers; the importance of test and training loss; how they should converge; under- and overfitting; how to fix these problems with specialized layers like dropout and batch normalization
  • Hands-on exercise: Play the third level of the deep learning game on how the convolve operation works internally; alter CNN layer parameters to fine-tune what features a model extracts with convolution; fine-tune your model; apply special layers to your model
  • Q&A

Break (5 minutes)

Text translation with recurrent networks (60 minutes)

  • Presentation: How sequences can be learned using recurrent networks; creating and presenting a new text translation example, sequence-to-sequence encoders; image captioning example
  • Hands-on exercise: Play the fourth level of the deep learning game on how the internals of recurrent network layers can learn; tune the hyperparameters of the model and do some basic training
  • Q&A