Generative Adversarial Network (GAN)

A dual-network architecture where a generator and discriminator compete to create hyper-realistic synthetic data.

Ian Goodfellow introduced GANs in 2014, fundamentally changing how machines handle creative tasks. The system pits two neural networks against each other: a generator creates fake samples (like human faces or textures) while a discriminator tries to detect the fraud. This zero-sum game forces the generator to produce high-fidelity outputs that are statistically indistinguishable from real-world data. From NVIDIA's StyleGAN generating non-existent people to medical researchers synthesizing rare pathology images, GANs provide the backbone for high-resolution image synthesis and complex data augmentation.

https://arxiv.org/abs/1406.2661