Discussing the article: "MQL5 Wizard Techniques you should know (Part 22): Conditional GANs"

[MetaQuotes]

Check out the new article: MQL5 Wizard Techniques you should know (Part 22): Conditional GANs.

Generative Adversarial Networks are a pairing of Neural Networks that train off of each other for more accurate results. We adopt the conditional type of these networks as we look to possible application in forecasting Financial time series within an Expert Signal Class.

Conditional Generative Adversarial Networks (cGAN) are a type of GAN that allow customization to the type of input data in their generative network. As can be seen from the shared link and in reading up on the subject, GANs are a pair of neural networks; a generator and a discriminator. Both get trained or train off of each other, with the generator improving at generating a target output while the discriminator is trained on identifying data (a.k.a. the fake data) from the generator.

The application of this is typically in image analysis where a generator network is used to come up with images and the discriminator network identifies whether the image it is fed with as input was either made up by the generator network or it is real. The training off each other happens by feeding the discriminator generator’s images alternated with real images and, like in any network, backpropagation would appropriately adjust the weights of the discriminator. The generator on the other hand, in non-conditional or typical settings is fed random input data and is supposed to come up with images that are as realistic as possible, regardless of this.

In a conditional GAN setting (cGAN) we do make a slight modification of feeding the generative network a certain type of data as input and not random data. This is applicable or useful in situations where thee type of data we feed to the discriminator is paired or is in 2 parts and the goal of the discriminator network is to tell if the input paired data is valid or made up.

Author: Stephen Njuki