Generative Adversarial Network (GAN)

Executive Definition: Generative Adversarial Network (GAN) — A machine learning framework where two neural networks—a generator and a discriminator—compete to produce realistic synthetic data.

The generator creates synthetic data, while the discriminator evaluates it against real data to detect fakes. Through this adversarial training process, the generator learns to produce increasingly realistic outputs. GANs are frequently used for image synthesis, super-resolution, and data augmentation.

Frequently Asked Questions

Are GANs still used now that diffusion models exist?

While diffusion models dominate image generation, GANs remain valuable for data augmentation, super-resolution, style transfer, and applications requiring real-time synthesis where diffusion models are too slow.

Why are GANs difficult to train?

The generator and discriminator networks must stay balanced. If the discriminator becomes too strong, the generator cannot learn; if too weak, the generator produces low-quality outputs. This instability requires careful hyperparameter tuning.

What are deepfakes?

Deepfakes use GAN technology and other AI methods to generate highly realistic, fabricated media. This raises significant ethical concerns regarding misinformation, consent, and identity theft.