What is Generative Adversarial Network (GAN) training instability and what are common solutions?

Machine Learning Hard

Machine Learning — Hard

What is Generative Adversarial Network (GAN) training instability and what are common solutions?

Key points

  • GAN instability stems from mode collapse and non-convergence
  • Wasserstein loss and gradient penalty are common solutions
  • Gradient saturation also contributes to GAN training instability
  • Minimax game dynamics play a role in GAN instability

Ready to go further?

Start full exam Practice more questions

Related questions