Artificial intelligence · 2020
Denoising diffusion (DDPM)
Ho, Jain and Abbeel make diffusion models match GANs on image quality.
Denoising Diffusion Probabilistic Models trained a U-Net to predict added noise, then generated images by iterative denoising. Stable, simple to train and diverse, diffusion quickly overtook GANs and became the engine of AI image and video generation.
What it built on 2
What it led to 5
Related measurements
The dashed line marks when it first worked.
Training compute of notable AI models Source: Epoch AI, Notable AI Models (CC BY 4.0)
ML hardware FLOP/s per dollar Source: Epoch AI, ML hardware (CC BY 4.0)