KAIST AIC313/CS378
Professor Sung · KAIST · 3 lectures with notes
Every lecture with its full transcript and chapters, and one search across all 3 lectures.
Lecture 14: Score Matching and Diffusion Models (KAIST CS479, Fall 2026)
DDPM noise prediction is score estimation, and multiscale Gaussian perturbations make score-based sampling work beyond high-density data regions.
Lecture 13: Denoising Diffusion Probabilistic Models 2 (KAIST CS479, Fall 2026)
DDPM training reduces to predicting noise or clean data from a randomly corrupted sample, with each reverse step refining the result.
Lecture 12: Denoising Diffusion Probabilistic Models 1 (KAIST CS479, Fall 2026)
DDPMs use a fixed forward noising process and a learned reverse process to generate diverse, high-quality data from Gaussian noise.