Diffusion Model is a popular Generative AI method.

Stable Diffusion and OpenAI Sora are diffusion models where diffusion takes place in latent space instead of image pixel space.

Video Contents:
00:38 Sampling from a Standard Gaussian Distribution
02:30 Forward Process
04:58 Noise Addition in Single Step
06:11 Variance Schedule
07:36 Reverse Process
08:32 Derivation of Variational Lower Bound
15:39 Computing L0 Term of VLB with Discrete Decoder
17:31 Derivation of Lsimple Minimizing L(t1) Term of VLB
23:49 Lsimple when t is 1 (Approximated Discrete Decoder)
25:56 Training
28:05 Sampling Using Trained Model
29:39 Learning Covariance

All images and animations in this video belong to me

References

Denoising Diffusion Probabilistic Models
Jonathan Ho, Ajay Jain, Pieter Abbeel
https://arxiv.org/abs/2006.11239

Improved Denoising Diffusion Probabilistic Models
Alex Nichol, Prafulla Dhariwal
https://arxiv.org/abs/2102.09672

Understanding Diffusion Models: A Unified Perspective
Calvin Luo
https://arxiv.org/abs/2208.11970

#machinelearning #computervision
#deeplearning #ai #aitutorial #education
#generativeai #diffusionmodels #ddpm
#stablediffusion #sora #openai
#variationallowerbound #evidencelowerbound
#loglikelihood #imageprocessing #datascience
#computervisionwithhuseyinozdemir