Week 7: The Generative AI Era & Building Diffusion Models

We spent six weeks classifying data. Now, it is time to create it. Welcome to the world of Generative AI. We explore the three architectures defining modern synthetic media: GANs, VAEs, and Diffusion Models. You will learn the math behind the minimax game, use the reparameterization trick to sample from latent space, and dive deep into the exact Markov chain mathematics powering state-of-the-art models like Midjourney and Sora. Core Topics: Generative Adversarial Networks (GANs), Minimax Loss & Mode Collapse, Conditional Generation (Pix2Pix, CycleGAN), Evaluating Generators (Fréchet Inception Distance), Variational Autoencoders (VAEs) & KL Divergence, Denoising Diffusion Probabilistic Models (DDPMs), Forward Noising, and Reverse Denoising. Week 7 Lab: Engineer the mathematical core of a Diffusion Model. Destroy clean 3D skeletal data with Gaussian noise, then write the denoising engine to hallucinate fluid human motion out of pure static.

8 weeks · 59 lectures · free to watch