Generative Modelling

A notebook series introducing the theory and implementation of generative models, from classical latent variable models to modern deep architectures. Each notebook contains mathematical derivations and from-scratch implementations.

📂 Repository Structure

1. Latent Variable Models

• 1.1 Mixture Models - model data as a mixture of distributions, introducing discrete latent variables to capture clusters.
• 1.2 Factor Analysis - explain data as linear combinations of continuous latent factors, uncovering hidden structure behind observed correlations.
• 1.3 Variational Autoencoders (VAEs) - introduce non-linear mappings with neural networks, using continuous latent variables and variational inference.

2. Flow-based Models

• 2.1 Discrete Flow Models (in progress)
• 2.2 Continuous Flow Models

3. Implicit Models

• 3.1 Generative Adversarial Networks (GANs)
• 3.2 Score-Based Models

4. Diffusion Models

• 4.1 Denoising Diffusion Probabilistic Models (DDPM) - generate data by iteratively denoising a simple distribution through a learned reverse-time Markov chain.

5. Autoregressive Models

Coming soon

📚 References

This series loosely follows the structure of Deep Generative Modeling by Jakub M. Tomczak (Springer, 2022) and Probabilistic Machine Learning: Advanced Topics, Generation by Kevin Murphy (MIT Press, 2023) alongside key seminal papers.