aifromphytsai/gan_sampling

Tutorial: A simple GAN to generate samples from Gaussian distribution

Tutorial: A simple GAN to generate samples from Gaussian distribution

PyTorch implementation. More details can be seen in the medium article here (mandarin version): https://medium.com/ai-academy-taiwan/sampling-by-gan-a-simple-case-study-6d0a8483592b

• Environment used:
  • torch 1.1.0
  • numpy 1.17.3
  • python 3.7.3

Learn sampling from 1D Gaussian distribution

Network architecture

• Generator

  • hidden layer: Fully-connected (32 nodes), ReLU activation
  • output layer: Fully-connected (1 node), no activation

• Discriminator

  • hidden layer: Fully-connected (32 nodes), ReLU activation
  • output layer: Fully-connected (1 node), Sigmoid activation

Results

• For mu = 0.0, sigma = 1.0:
  • After 1 epoch,
    
  • Afrer 2000 epochs,
    
  • Whole training evolution,
    

Further improvements

• minibatch feature
• WGAN implementation
• WGAN-GP implementation

References

1. https://github.com/togheppi/vanilla_GAN
2. https://github.com/ericjang/genadv_tutorial/blob/master/genadv1.ipynb
3. https://machinelearningmastery.com/how-to-develop-a-generative-adversarial-network-for-a-1-dimensional-function-from-scratch-in-keras/
4. https://blog.aylien.com/introduction-generative-adversarial-networks-code-tensorflow/
5. https://github.com/kremerj/gan
6. https://github.com/igul222/improved_wgan_training
7. https://github.com/AYLIEN/gan-intro/blob/master/gan.py