Domain Neighborhood

Generative Models

How models generate: likelihood, latent variables, diffusion/score models, flows, and the training tricks that make sampling work.

5 concepts5 published5 demos

Recommended Route

This sequence is ordered for learning rather than inventory: lower difficulty, fewer prerequisites, and more central concepts come first.

  1. 01
    Diffusion, Score-Based Models & Flow Matching

    Denoise noise into data: the diffusion forward process, score matching, and modern sampling via reverse-time dynamics and flow matching.

    22 mincodedemoafter Maximum Likelihood, Variational Autoencoders

    Check Maximum Likelihood first if the symbols feel slippery.

  2. 02
    Flow Matching & Rectified Flows

    Learn flow matching as velocity-label regression: straight conditional paths use x_t=(1-t)x0+t x1 but supervise the full target u_t=x1-x0.

    16 mincodedemoafter Diffusion, Score-Based Models & Flow Matching, Normalizing Flows: Tractable Density via Invertible Transforms

    Why this follows: Flow Matching & Rectified Flows uses Diffusion, Score-Based Models & Flow Matching directly.

  3. 03
    Normalizing Flows: Tractable Density via Invertible Transforms

    Invertible transforms trained with change-of-variables: tractable transformed densities and sampling, with architectural constraints to make the needed Jacobian determinants tractable.

    18 mincodedemoafter Maximum Likelihood, Variational Autoencoders

    Why this follows: both pages keep the generative models thread active.

  4. 04
    Score Matching & Score-Based Generative Models

    Learn the score field grad_x log p(x) without normalizing constants. Denoising score matching turns diffusion training into simple regression on noise.

    18 mincodedemoafter Maximum Likelihood, Diffusion, Score-Based Models & Flow Matching

    Why this follows: both pages keep the generative models thread active.

  5. 05
    Variational Autoencoders

    A latent-variable model trained by maximizing an evidence lower bound; the gap is KL(q_phi(z|x) || p_theta(z|x)), so the encoder is learned inference rather than just compression.

    20 mincodedemoafter Maximum Likelihood, Bayesian Inference, KL Divergence (Relative Entropy)

    Why this follows: both pages keep the generative models thread active.

All Published Notebooks

Browse the territory.

Advanced Bridges

Use these after the core path.