1. Overview of Generative Models

• Definition: Generative models are machine learning models that learn to generate data similar to a given dataset (e.g., images, text, or audio).
• Goal:
  • Learn the underlying data distribution.
  • Generate new samples that resemble the training data.

2. Generative Adversarial Networks (GANs)

How GANs Work

GANs involve two networks:

1. Generator (G):
   • Generates synthetic data (e.g., images) from random noise.
   • Tries to create data that is indistinguishable from real data.
2. Discriminator (D):
   • A binary classifier that distinguishes real data from fake data generated by G.

• Adversarial Process:
  • G: Maximizes the probability that D classifies its outputs as real.
  • D: Tries to minimize misclassifications by distinguishing real vs. fake data.

Loss Functions

1. Discriminator Loss:

2. Generator Loss:

• z: Latent variable sampled from a random distribution.

Applications of GANs

1. Image Generation:
   • Generate realistic images from random noise (e.g., faces, landscapes).
   • Example: StyleGAN creates high-quality human faces.
2. Image-to-Image Translation:
   • Transform images from one domain to another.
   • Examples:
     • Day-to-Night conversion.
     • Edges-to-Images.