1. Generative Model (Unsupervised Learning)

A Generative Model is an unsupervised learning approach that aims to model the underlying distribution of the training data. Once trained, it can generate new samples that are similar to the input data.

Examples:

• Generative Adversarial Networks (GANs)
• Variational Autoencoders (VAEs)
• PixelRNN, PixelCNN

Why use them? They can learn a meaningful representation of the data and can generate realistic new data samples.

2. PixelRNN – Explicit Density Modeling

PixelRNN is a generative model that sequentially generates pixels in an image. Each pixel is predicted based on all the previously generated pixels using RNNs.

• It models the joint probability of the image as a product of conditional distributions.
• Each pixel depends on the previously generated ones → slow due to sequential generation.

Pros:

• Accurate modeling of pixel dependencies.
• Explicit control over image generation.

Cons:

• Very slow because of sequential nature.
• Computationally expensive.

3. PixelCNN – CNN-based Generation