Map of Contents. Generative models that learn a data distribution and sample from it. Related: Machine Learning MOC for the neural network fundamentals, Natural Language Processing MOC for text generation and LLMs.
Foundational Generative Models
- Variational Autoencoders - Encoder-decoder with probabilistic latent space; enables image interpolation and semantic arithmetic via KL-regularized Gaussian prior.
- Generative Adversarial Networks - Adversarial two-network framework (Generator + Discriminator); produces sharper images than VAEs but is harder to train.
Image Generation
- Brief History & Model Types - Image generation has evolved rapidly over the past decade. GANs, VAEs, Diffusion models.
- List of Models - Current state-of-the-art image generation models and checkpoints.
Compute & Energy
- Energy Use of AI Models - Per-query and training energy, data centre demand compared with the flat 2010-2018 baseline, and projections to 2030.
Datasets & Sources
- Wikimedia Commons - Central repository for public domain and freely licensed media; a primary source for open-source visual datasets.