Diffusion Model
In machine learning,Diffusion models are also called diffusion probability models or score-based generative models.It is a class of latent variable generative models. Diffusion models consist of three main parts: the forward process, the backward process, and the sampling process. The goal of diffusion models is to learn the diffusion process that generates the probability distribution of a given dataset. They learn the latent structure of a dataset by modeling the way data points diffuse in the latent space.
In terms of computer vision, diffusion models can be applied to a variety of tasks including image denoising, inpainting, super-resolution, and image generation.They typically involve training a neural network to sequentially denoise images blurred with Gaussian noise. The model is trained to reverse the process of adding noise to an image. Once the training converges, it can be used for image generation, starting with an image composed of random noise and letting the network iteratively denoise it. OpenAI's text-to-image model DALL-E 2 is an example of this, which uses diffusion models for both the model's priors (generating image embeddings given a text caption) and the decoder that generates the final image. Diffusion models have recently found applications in natural language processing (NLP), particularly in areas such as text generation and summarization.Examples of common diffusion modeling frameworks used in computer vision include denoising diffusion probabilistic models, noise-conditioned scoring networks, and stochastic differential equations.
Diffusion models are inspired by nonequilibrium thermodynamics.They define a Markov chain of diffusion steps to slowly add random noise to the data, and then learn to reverse the diffusion process to construct the desired data samples from the noise. Unlike VAEs or flow models, diffusion models are learned via a fixed procedure and the latent variables have high dimensionality (the same as the original data).
References
【1】https://lilianweng.github.io/posts/2021-07-11-diffusion-models/