GAN (Generative Adversarial Network)
Generative Adversarial Network
A technology that pits two AIs against each other to generate realistic images and data.
In Simple Terms
GAN is a technology that pits a "generator" AI against a "discriminator" AI, having them compete to produce new data that looks just like the real thing. For example, it can create photos of people who don't actually exist, so realistic they're impossible to tell apart from real photos. It's also used in many other image processing tasks, such as sharpening blurry photos or automatically coloring line art.
Behind the Name
GAN is an acronym combining the English words Generative, Adversarial, and Network. It captures the idea of a network that improves its ability to create data by having two AIs compete like rivals.
Take a Closer Look!
GAN is a mechanism that generates new, realistic data by pitting two different AIs against each other.
Roughly speaking, it works by having a "generator," which creates new data, compete against a "discriminator," which tries to tell whether that data is real or fake.
Think of it like the relationship between a counterfeiter and a police officer.
The generator, playing the counterfeiter, tries to create fake bills convincing enough to fool the police.
Meanwhile, the discriminator, playing the police officer, keeps sharpening its skill at telling real bills from fake ones.
As these two AIs keep pushing each other to improve, the data they produce eventually becomes so realistic it's indistinguishable from the real thing.
This technology is useful for artificially generating training data needed to teach AI models, effectively padding out limited datasets. For example, it's used to recreate hard-to-collect data, such as medical images or rare defect cases, as a way to boost the overall learning accuracy of AI systems.
GAN is one of the training methods built on deep learning, the technology that lets computers automatically learn deep patterns in data on their own.
A key feature is that the two AIs compete against each other and sharpen their own accuracy, without needing humans to manually teach them the correct answers in detail.