Score-Based Diffusion via SDEs in 2D

Implementation of a variance-preserving diffusion model using stochastic differential equations (VP-SDEs) in PyTorch.

Features

• Forward diffusion process
• Time-dependent score network
• Reverse-time SDE sampling
• Noise-conditional classifier
• Classifier guidance
• 2D mixture-of-Gaussians dataset

Concepts

• Score matching
• VP-SDE
• Euler-Maruyama integration
• Fourier time embeddings
• Classifier guidance

Credits

Code from assignment, partially authored by me.