AAI Handbook
← All models
architecture

Diffusion Model

advancedself_supervised
parametricgenerationrepresentationimageaudiovideo

A diffusion model learns to turn noise into structured data by repeatedly reversing a gradual corruption process.

Mechanisms

iterative denoisingnoise predictionbackpropagation

Properties

generativerepresentation learningstochastic generation

Constraints

high computehigh memoryrequires large dataslow inference

Practical profile

Explainability
low
Training cost
high
Inference cost
high
Data appetite
high
Concepts to understand
Backpropagation