In this course, students explore the theoretical and mathematical framework of GANs and experience hands-on guided workshops and practical applications in the data synthesis space. We focus on image synthesis while learning the fundamental concepts, mathematical formulation, and practical aspects of building and training generative models required to create innovative AI systems. This course will also introduce ChatGPT-like systems.

GANs are one of the most valuable components of modern AI, be it computer vision, anomaly detection, or language models. In language modeling alone, GANs and other neural network techniques have shown remarkable milestones for text generation, text summarization, language translation, question answering, and more. The recent success of DALL-E 2 and ChatGPT highlight these accomplishments.

ChatGPT leverages the GANs framework and facilitates text-based conversation with humans by generating, synthesizing, and allowing interactive responses to questions. ChatGPT engages in communication just like another human and is just the beginning, with many more excellent AI applications on the horizon. This GANs course teaches you the fundamentals required to build such innovative AI systems and lays a path to the future.

Working in a research environment, you’ll learn the problems and challenges associated with GANs and overcome them at the production level. Students will perform image translation and synthesis tasks with state-of-the-art networks, such as Pix2Pix and CycleGAN. You’ll implement deep learning algorithms from technical papers for deep generative models and focus on building an intuition of efficient training of DL and GAN models.

Learning Outcomes
At the conclusion of the course, you should be able to:

• Define the generative models and their use cases
• Build the generative adversarial networks (GANs) framework
• Train GANs with variants of their cost functions
• Explain the generator, discriminator, GAN loss function, and adversarial loss
• Build conditional GANs and WGAN
• Perform image translation and synthesis tasks with state-of-the-art networks, such as Pix2Pix, CycleGAN

Requisite Knowledge
You need to be familiar with probability theory and linear algebra, programming, and deep learning.