M10

Question 1

GAN (Generative Adversarial Network)

Answer 1

• made of two competing networks, each trying to beat the other
  
  • discriminator (determines whether an image is real or fake)
  • generator (generates fake images)

Question 2

Adversarial machine learning

Answer 2

Technique in machine learning to fool a model by giving it “malicious” input. Not fixed by reducing overfitting.

Question 3

Types of adversarial attack

Answer 3

• Fast gradient sign method
• Better attack methods
• Black box attack

Question 4

Types of GAN

Answer 4

• Deep Convolutional GAN (DC GAN)
• Conditional GAN (CGAN)
• Least Square GAN (LSGAN)
• Auxiliary Classifier GAN (ACGAN)
• Dual Video Discriminator GAN (DVD GAN)
• SRGAN
• Cycle GAN
• Info GAN