stem/AI/Neural Networks/CNN/GAN/GAN.md

Fully Convolutional

• Remove Max Pooling
  • Use strided upconv
• Remove MLP layers
  • Hurts convergence in non-classification
• Normalisation tricks
  • Batch normalisation
    • Batches of 0 mean and variance 1
  • Leaky Activation Functions#ReLu

Stages

Generator, G

• Synthesise 'fake' images
• From noise

Discriminator, D

• Discriminator is a classifier
  • Is image fake or real

! !

!

Training

! !

Code Vector Math for Control

!

• Do Interpretation#Activation Maximisation to derive code for an image !