24 May 2023
GANs were developed in June 2014 by Ian Good fellow and colleagues.
GANs involve two neural networks competing in a zero-sum game, where one network's gain is another's loss.
GANs require training data to learn from and generate new data.
Well-trained GANs can generate data that closely resembles the training data, like realistic human faces.
GAN training is a minimax game between the generator (G) and discriminator (D) networks, seeking a balanced equilibrium.
GANs aim to recover the training data distribution.
GANs find applications in image generation, text-to-image synthesis, style transfer, and video synthesis.
GANs use an adversarial loss function to encourage generator output resembling real data.
It faces challenges like mode collapse & learning instability. Research aims to improve training techniques.
It incorporates additional information to influence data generation based on specific conditions.