concept
The diffusion model DVAE (Diffusion Variational Autoencoder) is a generative model that combines a variational autoencoder (VAE) and a diffusion process. It combines the concepts of time-continuous diffusion processes and variational autoencoders for modeling and generating data with time series structure, such as videos, time series, etc.
Main concepts:
-
Variational Autoencoder (VAE): VAE is a generative model that generates new samples by learning the underlying distribution of the data. It consists of an encoder that maps the input data to distribution parameters in the latent space, and a decoder that maps the latent variables back to the original data space.
-
Diffusion process: The diffusion process is a random process that is continuous in time and describes the random walking behavior of random particles in space. In DVAE, this process is used to simulate the data generation process and is combined with VAE.
-
Model structure: The structure of DVAE includes the modeling of the diffusion process and the learning of latent variables based on variational autoencoders. It combines a time-continuous diffusion process and a stochastic encoder-decoder structure.
working principle:
-
Modeling of diffusion models: DVAE treats the data as a diffusion process and models the data in terms of time steps. This process is different from the traditional discrete time step model, which treats data changes as a continuous time evolution process.
-
Learning of latent space: DVAE uses variational autoencoders to learn a latent space representation of the data. This latent space can capture the data generation process and be able to generate new samples.