Diffusion Model (DDPM)

Jonathan Ho, Ajay Jain & Pieter Abbeel, 2020

Denoising Diffusion Probabilistic Models (DDPM), introduced by Ho, Jain, and Abbeel in 2020, generate data by reversing a Markov chain that gradually adds Gaussian noise. Starting from pure noise (t=T), a trained neural network predicts and removes noise at each step, progressively clarifying the signal until a clean sample emerges (t=0). The visualization shows a grid transitioning from random noise to a recognizable circle pattern, with a step counter, noise level bar, and timeline thumbnails illustrating key stages of the denoising process.