Three graduate assignments on generative modelling, each a self-contained, executed Jupyter notebook with derivations, implementation and results.
| Notebook | Topics |
|---|---|
HW1/submission/ee048954_hw1_id1_id2.ipynb |
Kernel density estimation, normalizing flows (affine coupling / RealNVP-style), likelihood evaluation |
HW2/ee048954_hw2.ipynb |
Langevin dynamics sampling, energy-based models trained with contrastive divergence, denoising noisy digits |
HW3/ee048954_hw3.ipynb |
Score matching, denoising score matching, annealed Langevin sampling |
All notebooks run on a single GPU with PyTorch, NumPy and Matplotlib. Model checkpoints and intermediate tensors are not tracked.