Diffusion Models

Diffusion models are a type of generative model that uses a diffusion process to generate data.

History

• Diffusion models were first introduced in the 1990s by Sepp Hochreiter and Jürgen Schmidhuber.

• They were initially developed for speech recognition tasks, but have since been applied to a wide range of sequential data tasks, including natural language processing, music generation, and machine translation.

How they work:

Diffusion models are a type of generative model that uses a diffusion process to generate data.

Let's consider a simple example of how a diffusion model might work for image generation:

1. Start with random noise: Begin with a completely random image.

2. Gradual denoising: Slowly remove noise from the image over multiple steps.

3. Learned reverse process: The model learns to reverse this process, gradually adding structure to the noise.