Generative AI & Foundation Models
Master modern generative models — from diffusion to large language models and multimodal AI. Dive into the cutting edge of AI.
Level: Advanced · Category: Generative AI · Estimated time: 7 hours
Prerequisites
- Transformers & NLP
- Computer Vision with Deep Learning
Lessons
- Variational Autoencoders (VAEs) — Latent spaces, the reparameterization trick, and generative modeling.
- GANs Deep Dive — GAN training dynamics, mode collapse, Wasserstein GANs, and StyleGAN.
- Diffusion Models — DDPM, score matching, classifier-free guidance, and Stable Diffusion.
- Scaling Laws & Emergent Abilities — Chinchilla scaling, compute-optimal training, and emergent properties of large models.
- LoRA & Parameter-Efficient Fine-Tuning — LoRA, QLoRA, adapters, and fine-tuning large models on consumer hardware.
- Multimodal Models — CLIP, DALL-E, GPT-4V, and models that understand both text and images.
- Building Applications with LLMs — RAG, agents, tool use, LangChain, and production LLM applications.
Topics covered
generative-ai, diffusion, llm, lora, multimodal, stable-diffusion, foundation-models