Understanding Generative Adversarial Networks (GANs)

Glossary/G/Generative Adversarial Networks (GANs)

What are Generative Adversarial Networks (GANs)?

Generative Adversarial Networks (GANs) are a class of deep learning models that consist of two neural networks: a generator and a discriminator. GANs are designed to generate synthetic data that resembles real data by learning from a training dataset. The generator network generates new samples, while the discriminator network evaluates the authenticity of the generated samples compared to the real ones.

The generator network takes random input, often referred to as noise or a latent vector, and maps it to the data space. It learns to generate samples that mimic the patterns and distribution of the training data. The goal of the generator is to produce samples that are indistinguishable from real data.

Related terms

Deep Learning

Deep Learning is a subfield of Machine Learning (ML) that focuses on developing and training artific...

Neural Network

A neural network is a fundamental concept in the field of artificial intelligence and machine learni...

Not to be confused with:

Deep Learning

Deep Learning is a subfield of Machine Learning (ML) that focuses on developing and training artific...

Neural Network

A neural network is a fundamental concept in the field of artificial intelligence and machine learni...

Deep Learning

Deep Learning is a subfield of Machine Learning (ML) that focuses on developing and training artific...

Neural Network

A neural network is a fundamental concept in the field of artificial intelligence and machine learni...

Deep Learning

Deep Learning is a subfield of Machine Learning (ML) that focuses on developing and training artific...

Neural Network

A neural network is a fundamental concept in the field of artificial intelligence and machine learni...

On the other hand, the discriminator network is trained to classify whether a given sample is real or generated by the generator. It acts as an adversary to the generator, aiming to correctly differentiate between real and fake samples. The discriminator provides feedback to the generator, guiding it to produce better and more realistic samples.

During training, the generator and discriminator networks play a two-player minimax game. The generator tries to fool the discriminator by generating samples that the discriminator cannot distinguish from real ones. Simultaneously, the discriminator tries to improve its ability to correctly classify the samples. Through this adversarial process, both networks learn and improve iteratively.

The iterative nature of GAN training often involves a delicate balance. If the generator becomes too good at generating realistic samples, the discriminator may struggle to differentiate them, leading to a stalemate. On the other hand, if the discriminator becomes too powerful, it may easily detect even subtle differences between real and generated samples, hindering the generator's progress. Achieving the right equilibrium between the generator and discriminator is a critical aspect of GAN training.

Back to glossary