GANs Loving: Intro to Generative Adversarial Networks

The course

Three module hands on video course that will get you coding and understanding Generative Adversarial Networks (GANs). GANs are what AI pioneer and director of AI at Facebook Yann LeCun called .

Module 1: GANs basic concepts and coding a simple GAN

Module 2: Generative models overview and coding a DCGAN

Module 3: CycleGAN and GANs evaluation

Each module comes with a Google Colab notebook for you to work along in, and a series of exercises and readings for you to complete to lock in your learning.

About the course

This course was first taught in August 2019 to a group of 30 in-person and online learners. Student response to the course was overwhelmingly positive, so we've continued teaching it. Some of the first students went on to write an extensive blog post on modern GANs as their capstone project. Past student review responses are shown below.

Who is this course for?

If you're a digital marketer who's never programmed before, this course may not be for you.

However, this course has been beneficial for people with the following backgrounds.

• Data scientists looking to acquire new skills
• Developers looking to acquire AI experience
• Data engineers looking to learn more about data science
• Analysts looking for a career transition into data science
• Civil/Mech engineers looking for career transition into tech
• New grads looking to differentiate themselves from their peers
• Artists looking to create generative art
• People who are just genuinely curious about such a cool technology

About the instructor

Andrew has worked as a data scientist for almost 10 years. He's built planning models for million acre forests, scheduling models for airlines, and worked extensively with generative AI models.

Of these generative AI models, GANs are his favourite, and the ones he's worked with the most. At Looka he built a 5 person data team from scratch that used GANs and other generative models to generate beautiful original graphic designs. He currently serves as the CTO of Looka.

Andrew has a B.Sc in mathematics and an M.A.Sc in Industrial Engineering.

Feel free to connect on LinkedIn.

Selected videos of Andrew speaking about GANs

Intro video for this course.

Andrew speaking about GANs in 2018.

Andrew giving an overview of GANs in 2019.

About the models we'll code

Basic GAN

This is the model that Ian Goodfellow introduced in his seminal 2014 paper that put forth the concept of generative adversarial nets. Coding this model helps the student understand the fundamentals of GANs.

Deep Convolutional GAN (DCGAN)

This is the model that first showed the world that GANs were more than just a cool concept, but could be used to generate surprisingly high resolution images. Coding this model introduces the student to more advanced concepts and fleshes out their understanding.

CycleGAN

This unpaired image to image translation model tests the student's understanding of all the concepts so far. It is also the most versatile GAN we'll look at. It's capable of being used for mapping, photo editing, and style transfer, among many other uses.

Pre-requisites

• Experience with Python
• Computer with internet connection
• Experience with PyTorch is helpful though you can learn as you go
• Same goes for neural networks

Contact

Direct any questions to andrew.brownmartin at gmail dot com