GAN (Generative Adversarial Network)

A GAN (Generative Adversarial Network) is a machine learning setup used to generate highly realistic synthetic data (e.g., images, videos, deepfakes).

Core idea:

• It uses two neural networks:
  • Generator: creates fake data from noise.
  • Discriminator: evaluates data and tries to detect what’s fake vs real.
• They are trained in competition: the generator tries to fool the discriminator; the discriminator tries not to be fooled.
• This adversarial process pushes the generator to produce increasingly realistic outputs with little human fine-tuning.

GANs are especially powerful for narrow, specific tasks like realistic image or video generation, rather than broad, general-purpose AI.