Generative Adversarial Networks (GANs)

A GAN consists of two networks; a generator (G) and a discriminator (D), given a set of training examples, G will generate outputs and D will classify them as either being from the same distribution as the training examples or not. In doing so D is optimized so as to be able to discriminate between examples from the training example and from the generator network which in turn is optimized to fool D into classifying its output as being drawn from the training examples. After such training G can now generate samples with properties very similar to those of the training examples. GANs tend to be devilishly hard to train

See AI/Generative AI/GenAI

Resources

Courses

Talks

Code

References

Subtopics

GANs for super-resolution

See AI/Computer Vision/Super-resolution#GAN-based

GANs for missing data, imputation and inpainting

See AI/Computer Vision/Inpainting and restoration#GAN-based

Image-to-image translation. Conditional GANs

See AI/Computer Vision/Image-to-image translation#GAN-based

GANs for spatio-temporal data generation

GANs for representation learning and image synthesis

Semi-supervised GANs

Few/one-shot learning GANs

See AI/One, few-shot learning#Few/one-shot learning GANs

GANs for anomaly detection