2025/Fall/Sharif University of Technology
- Homework 1: Deep Autoregressive Models
- Theory: Gaussian properties, AR models, Real NADE parameters.
- Practical: WaveNet for audio and PixelCNN/PixelRNN for images.
- Homework 2: Variational Autoencoders (VAEs)
- Theory: CVAE derivation, Cauchy–Schwarz divergence, posterior collapse.
- Practical: Probabilistic graph forecasting and CVAE for MNIST.
- Homework 3: Normalizing Flows
- Theory: 1x1 Convolutions in Glow, Continuous Normalizing Flows, MAF vs IAF.
- Practical: Building flows from scratch and image inpainting with Glow.
- Homework 4: Generative Adversarial Networks (GANs)
- Theory: Divergence minimization, Wasserstein GAN, f-GAN, and AC-GAN.
- Homework 5: Energy-Based & Score-Based Models
- Theory: MCMC ergodicity, score-matching variants (ESM, ISM, DSM), and EBMs.
- Practical: Implementation of Score-based models.