Description

Deep Learning: GANs and Variational Autoencoders

Generative AI is transforming entertainment, design, and science. In this course, you’ll build two of the most powerful generative models: **Generative Adversarial Networks (GANs)** and **Variational Autoencoders (VAEs)**—from theory to implementation in TensorFlow and PyTorch.

Projects You’ll Build

  • A GAN that generates new human faces (like ThisPersonDoesNotExist)
  • A VAE that learns a compressed latent space of fashion images
  • An age progression GAN that turns young faces into elderly ones
  • A denoising autoencoder that cleans corrupted images

Key Concepts Covered

  • GAN architecture—generator vs. discriminator, adversarial training
  • VAE theory—probabilistic encoder, reparameterization trick, KL divergence
  • Training tricks—label smoothing, spectral normalization, Wasserstein loss
  • Evaluation metrics—Inception Score, FID, visual inspection
  • Advanced variants—DCGAN, cGAN, InfoGAN, β-VAE

Why Generative Models?

  • Data augmentation—generate synthetic training data
  • Art and design—AI-generated fashion, music, and visuals
  • Anomaly detection—VAEs identify outliers by reconstruction error
  • Foundation for diffusion models—understand the evolution of generative AI

Who Should Take This?

  • Deep learning enthusiasts wanting to go beyond classification
  • Computer vision engineers exploring generative applications
  • Students building advanced capstone projects
  • Researchers needing a practical intro to generative modeling

From Pixels to Imagination

You’ll graduate with the ability to **create**, not just classify—unlocking a new dimension of AI.

Ready to make AI that creates? Enroll now.