How GANs are the opposite of discriminative algorithms (ie. generative)

Paenuth

Neophyte
Joined
May 18, 2019
Messages
2
Reaction score
1
Location
Sri Lanka
I've been working with GANs, and as an introduction into the concept of what they are, it's easier to compare these algorithms to typical discriminative algos.

"Discriminative" means that the algorithm tried to categorize data based on its features. Given a certain feature or features within the instance data, what is its classification label? A good example of this is algorithms which have thousands of images of moles passed into them with the aim of predicting which images contain malignant skin cancers and which do not.

GANs approach problems from the opposite direction. Assuming that a mole is cancerous, how likely is each feature to be present? This is only a very basic explanation of the difference, but one that is important to understand.
 

Alframen

Neophyte
Joined
May 19, 2019
Messages
4
Reaction score
1
Location
Jordan
Thanks for your insight. Do you work only with GANs or with other models as well? Do you think there is more potential for GANs in the future, or for CNNs, or RNNs?