Порождающие модели (теория и практика, Р.В. Исаченко, В.В. Стрижов)/Группа 674, осень 2020
Материал из MachineLearning.
(Различия между версиями)
(→Программа курса) |
(→Программа курса) |
||
(4 промежуточные версии не показаны) | |||
Строка 65: | Строка 65: | ||
| lecture 9 | | lecture 9 | ||
| 28.10 | | 28.10 | ||
- | | Wasserstein GAN. Spectral Normalization GAN. f-divergence | + | | Wasserstein GAN. Spectral Normalization GAN. f-divergence. |
| [https://github.com/r-isachenko/2020-DGM-course/blob/master/lectures/lecture9/Isachenko2020DeepGenerativeModels9.pdf slides] | | [https://github.com/r-isachenko/2020-DGM-course/blob/master/lectures/lecture9/Isachenko2020DeepGenerativeModels9.pdf slides] | ||
- | | [video] | + | | [https://youtu.be/tc-YZ_pjwWo video] |
+ | <!-- Конец занятия --> | ||
+ | |- <!-- Новое занятие --> | ||
+ | | lecture 10 | ||
+ | | 11.11 | ||
+ | | GAN evaluation. Advanced GANs (SAGAN, BigGAN, ProGAN, StyleGAN). | ||
+ | | [https://github.com/r-isachenko/2020-DGM-course/blob/master/lectures/lecture10/Isachenko2020DeepGenerativeModels10.pdf slides] | ||
+ | | [https://youtu.be/kKmQlLOQLIY video] | ||
+ | <!-- Конец занятия --> | ||
+ | |- <!-- Новое занятие --> | ||
+ | | lecture 11 | ||
+ | | 25.11 | ||
+ | | Disentanglement (InfoGAN, beta-VAE, DIP-VAE, FactorVAE). | ||
+ | | [https://github.com/r-isachenko/2020-DGM-course/blob/master/lectures/lecture11/Isachenko2020DeepGenerativeModels11.pdf slides] | ||
+ | | [https://youtu.be/xqP01Q4jLIM video] | ||
+ | <!-- Конец занятия --> | ||
+ | |- <!-- Новое занятие --> | ||
+ | | lecture 12 | ||
+ | | 09.12 | ||
+ | | Continious-in-time models (NeuralODE, FFjord). Quantized latent models (VQ-VAE, VQ-VAE-2, FQ-GAN). | ||
+ | | [https://github.com/r-isachenko/2020-DGM-course/blob/master/lectures/lecture12/Isachenko2020DeepGenerativeModels12.pdf slides] | ||
+ | | [https://youtu.be/K0qu8Is-i94 video] | ||
<!-- Конец занятия --> | <!-- Конец занятия --> | ||
|} | |} | ||
Строка 93: | Строка 114: | ||
| 26.10 | | 26.10 | ||
<!-- Конец занятия --> | <!-- Конец занятия --> | ||
- | | | + | |- <!-- Новое занятие --> |
+ | | homework 4 | ||
+ | | GAN. WGAN. | ||
+ | | [https://github.com/r-isachenko/2020-DGM-course/blob/master/homeworks/homework4/hw4.ipynb link] | ||
+ | | 09.11 | ||
+ | <!-- Конец занятия --> | ||
|} | |} | ||
Текущая версия
Программа курса
№ | Дата | Тема | Слайды | Видео |
---|---|---|---|---|
lecture 1 | 02.09 | Logistics. Motivation. Autoregressive models (MADE, WaveNet, PicelCNN). | slides | video |
lecture 2 | 09.09 | Bayesian framework. Latent variable models. EM-algorithm. | slides | video |
lecture 3 | 16.09 | EM-algorithm. VAE. Mean field approximation. | slides | video |
lecture 4 | 23.09 | Flow models (NICE, RealNVP, RevNet, i-RevNet). | slides | video |
lecture 5 | 30.09 | Flow models (Glow, Flow++). Flows in VAE. Autoregressive flows (IAF). | slides | video |
lecture 6 | 07.10 | Autoregressive flows (IAF, MAF, Parallel WaveNet). ELBO surgery. | slides | video |
lecture 7 | 14.10 | VampPrior. Posterior collapse (PixelVAE, VLAE). Decoder weakening. IWAE. | slides | video |
lecture 8 | 21.10 | Vanila GAN. Vanishing gradients, mode collapse. KL vs JSD. DCGAN. Wasserstein distance. | slides | video |
lecture 9 | 28.10 | Wasserstein GAN. Spectral Normalization GAN. f-divergence. | slides | video |
lecture 10 | 11.11 | GAN evaluation. Advanced GANs (SAGAN, BigGAN, ProGAN, StyleGAN). | slides | video |
lecture 11 | 25.11 | Disentanglement (InfoGAN, beta-VAE, DIP-VAE, FactorVAE). | slides | video |
lecture 12 | 09.12 | Continious-in-time models (NeuralODE, FFjord). Quantized latent models (VQ-VAE, VQ-VAE-2, FQ-GAN). | slides | video |
Домашние задания
№ | Тема | Ссылка | Дедлайн |
---|---|---|---|
homework 1 | Autoregressive models. | link | 28.09 |
homework 2 | Latent variable models. Flows. | link | 12.10 |
homework 3 | Autoregressive flows. Flows in VAE. | link | 26.10 |
homework 4 | GAN. WGAN. | link | 09.11 |
Полезные ссылки
Короткая ссылка на страницу: https://bit.ly/3i3N4G0
Видеолекции: link
Отзывы о курсе: link
Репозиторий курса на github: link