🗓️ Week 13 - Deep Generative Models

Author

Affiliation

Dr. Simone Rossi

EURECOM

This week we explore deep generative models that combine latent variable models with neural networks. We cover variational autoencoders (VAEs), including the evidence lower bound (ELBO), the reparameterization trick, and amortized inference. We then move to likelihood-free generative models, starting with generative adversarial networks (GANs) and their minimax training objective. Finally, we introduce diffusion models as a state-of-the-art approach to generative modeling based on gradually denoising data.

🖥️ Lecture Slides

• Deep Generative Models
• Likelihood-Free Generative Models

📑 Lecture Notes (PDF)

• Deep Generative Models
• Likelihood-Free Generative Models

📚 Recommended Reading

• Murphy (2023): Chapter 21 (Variational Autoencoders)
• Murphy (2023): Chapter 26 (Generative Adversarial Networks)
• Murphy (2023): Chapter 25 (Diffusion Models)

📙 References

Murphy, K. P. (2023). Probabilistic machine learning: Advanced topics. MIT Press. http://probml.github.io/book2