Course contentsShow
← Machine Learning and Deep Learning
Lesson 1469 of 3,53833. Generative Adversarial NetworksFree lesson

What GANs Are and Why They Matter

Overview of generative adversarial networks as a framework for training generative models through adversarial competition.

What GANs Are and Why They Matter

You'll learn what generative adversarial networks are and why they revolutionized generative modeling.

The Core Idea

A Generative Adversarial Network (GAN) is a framework for training generative models through a game between two neural networks: a generator and a discriminator. Unlike VAEs that learn to encode and decode data, GANs learn to generate new data by pitting two models against each other in adversarial competition.

Think of it like an art forger (generator) trying to create fake paintings while an art detective (discriminator) tries to spot the fakes. As the detective gets better at spotting forgeries, the forger must improve their technique. As the forger gets better, the detective must sharpen their skills. This back-and-forth competition pushes both to excel—and the result is a generator that creates remarkably realistic outputs.

Why GANs Matter

GANs introduced a fundamentally different approach to generative modeling. Instead of explicitly modeling probability distributions (like VAEs do with the ELBO), GANs learn implicitly through competition. This adversarial training often produces sharper, more realistic images because the discriminator forces the generator to match the real data distribution precisely.

The impact has been enormous: GANs enabled photorealistic face generation, artistic style transfer, image super-resolution, and countless creative applications that seemed impossible before 2014.

The Key Difference

While VAEs optimize a mathematical objective (reconstruction + KL divergence), GANs play a game. The generator tries to fool the discriminator, and the discriminator tries not to be fooled. This game-theoretic setup creates a unique training dynamic that we'll explore in depth.

Key Takeaway: GANs are a framework where two neural networks compete—a generator creating fake data and a discriminator judging real vs. fake—driving both to improve and enabling powerful generative modeling.