Basic Understanding of Generative Adversarial Network (GAN)

Generative Adversarial Network

Generative Adversarial Network (GAN)

A Generative Adversarial Network (GAN) is a class of machine learning frameworks designed by Ian Goodfellow and his colleagues in June 2014.

The idea behind is this:

1. By Generator (G) randomly generates new data set with the same features or statistics as the training set.
2. Consider random samples from both newly generated and original data sets as Db’ and Db respectively
3. For each newly generated sample (Db’), generate Yi=1 (class label of features as 1), and for original sample (Db), generate Yi=0 (class label of features as 0).
4. Pass both Db’ and Db to Model (M) (let say simple classification model), then check the error or loss (let log loss for binary classification).
5. If the error or loss is high then your Generator is working good. Means your generated data and original data is similar.

Generative Adversarial Network (GAN)

The Generative Adversarial Network (GAN) name is because Generative means generating new data, Adversarial means the Model (M) checks similarity between the both Db’ and Db. And Network means, Here in G and M we use Deep Neural Network or CNN’s.