made a GAN summary for ECG project

1. GAN summary
Liangqun Lu
2018-07-13
*Most slides are from Hung-yi Lee ‘Generative Adversarial Network
(GAN), 2018’, videos and slides

2. Outline
● GAN introduction
○ GAN theory
○ Issues and Solutions
● Examples from GAN Zoo
○ WGAN and WGAN-GP
○ EBGAN
○ LSGAN
○ DCGAN
● Feature extraction-based GANs
○ Conditional generation
○ Unsupervised conditional generation

3. ● Discriminator learns to assign high scores to real objects and low scores to
generated objects.
● Generator learns to “fool” the discriminator

14. GAN issues and solutions
● JS convergence
● Wasserstein distance
● Using an autoencoder as discriminator D
● Mode Collapse and Dropping
● GAN ensemble
● Inception score

32. DCGAN
Source: https://www.slideshare.net/shyamkkhadka/unsupervised-learning-represenation-with-dcgan