Syllabus

Date	Topics	Video from playlist
April 3^rd, 2026	[slides] Lecture 1: Diffusion • Background on vision • Motivation behind diffusion • Diffusion in DDPM • Training derivation, ELBO • Inference • Faster sampling with DDIM	1:46:26
April 10^th, 2026	[slides] Lecture 2: Score matching • Motivation behind score matching • Score estimation • Denoising score matching • SDE formulation • Training, inference • Probability flows • Parallel with diffusion	1:48:48
April 17^th, 2026	[slides] Lecture 3: Flow matching • Motivation behind flows • History on flows • Conditional flow matching • Training, inference • Rectified flow • Parallel with diffusion and score matching	1:47:34
April 24^th, 2026	[slides] Lecture 4: Latent space and guidance • Variational autoencoders • Latent diffusion models • Text representation • Image representation • Contrastive learning, CLIP, SigLIP • Guidance (classifier-based, classifier-free)	1:40:58
May 1^st, 2026	Midterm [exam] [solutions]
May 8^th, 2026	[slides] Lecture 5: Image generation architectures • Convolutions • U-Net • Attention mechanism • Diffusion Transformers • Multimodal DiT • Optimizations	1:46:26
May 15^th, 2026	[slides] Lecture 6: Model training • Training lifecycle • Pretraining, curriculum learning • Continued training, supervised finetuning • Preference tuning with Flow-GRPO, Diffusion-DPO • Tuning with DreamBooth, LoRA • Distillation	1:40:58
May 22^nd, 2026	[slides] Lecture 7: Evaluation • Human ratings (pointwise, pairwise) • Elo rating system • Reference-free metrics (FID, CLIPScore, PickScore) • Reference-based metrics (MSE, PSNR, SSIM, LPIPS) • Multimodal LLMs • MLLM-as-a-Judge	1:41:12
May 29^th, 2026	[slides] Lecture 8: Trending topics • Class recap • State-of-the-art models • Extension from image to video • Alternatives to generation • Parallel with the text world • Conclusion	1:49:34
June 8^th, 2026	Final [exam] [solutions]