Diffusion Models Study Plan

Generative vs discriminative models, taxonomy of generative approaches, likelihood-based models, latent variables, evaluation metrics, and why diffusion models.

Gaussian distributions, Markov chains, KL divergence, variational inference, stochastic differential equations, and Langevin dynamics.

Adding noise to data, noise schedules, the forward SDE, reparameterization trick, signal-to-noise ratio, and continuous vs discrete time.

Score functions, denoising score matching, the reverse SDE, DDPM training objective, parameterizing the noise predictor, and loss weighting.

Ancestral sampling, DDIM deterministic sampling, DPM-Solver, progressive distillation, consistency models, and guidance during sampling.

Autoencoders for latent space, LDM architecture, cross-attention conditioning, text encoders, the Stable Diffusion pipeline, and ControlNet.

Classifier guidance, classifier-free guidance, text-to-image systems, image-to-image translation, inpainting, and super-resolution.

Flow matching, rectified flows, video diffusion, 3D generation, Diffusion Transformers (DiT), deployment optimization, and ethical considerations.