Diffusion Models (DDPM, DDIM, Stable Diffusion, U-Net) Deep Dive By Google Engineer | Multimodal LLM скачать в хорошем качестве

Diffusion Models (DDPM, DDIM, Stable Diffusion, U-Net) Deep Dive By Google Engineer | Multimodal LLM

Diffusion models are widely used in most of the popular multimodal LLM today, including chatgpt, gemini etc, and is a key technical driver behind the latest wave of generative AI applications you're seeing across the tech industry (for example midjourney, stable diffusion). As a regular normal SWE, I'd like to share with you topics including DDPM, DDIM, guided diffusion, latent diffusion, U-Net architecture. I’ll not only focus on the high levels, but also the training and sampling details, model architectures and underlying math. Related videos: AutoEncoders Deep Dive:    • AutoEncoders Deep Dive By Google Engineer ...   Generative Image Model Overview:    • Generative Image Models (Diffusion, GAN, V...   Transformer Deep Dive:    • Transformer Deep Dive with Google Engineer...   #ai #diffusion #stablediffusion #midjourney #llm #google #multimodal #maths 0:00 Intro 1:56 Diffusion Background 4:53 Forward Diffusion 9:06 Reverse Diffusion 15:02 Loss Function 18:54: Why Is Scheduler Needed? 20:29 Why Predict Noise? 22:12 Training Process 25:02 Sampling Process 27:09 DDIM 33:30 U-Net Architecture 39:10 Guided Diffusion 40:40 Stable Diffusion