We kicked off this new AI-focused series by Cyber expert Ravi Das with an article about the Variational Autoencoder (VAE). In this second article, he moves on to a topic (or at least a term) that may be more familiar to you: the General Adversarial Network, or GAN. 

In the world of Artificial Intelligence (AI), the General Adversarial Network (GAN) is a Machine Learning Model of sorts, in which new datasets can be created from what it has previously learned. However, the GAN is not deemed to be as sophisticated as the Variational Autoencoder (VAE), as it can only produce newer kinds of datasets from the sources that it has been fed information and data from. In contrast, the VAE, due to the extra number of components it possesses, can produce newer types of datasets from other sources of information and data that it has not previously trained upon. 

The GAN consists of four primary components, which are as follows:

  • The Generator: Although limited in sophistication to the “Encoder” (reviewed in the last article), the Generator still actually produce newer types and kinds of Datasets which closely parallel the Datasets that have been fed previously into the Generative AI Model. However, it should be noted here that at the initial outset, the Datasets that have been produced by the “Generator” may not closely match the Datasets that the Generative AI Model has been fed and trained upon.
  • The Discriminator: In technical terms, this component of the GAN is also known as the “Binary Classifier,” in the sense that it can take the “real” Datasets that have been ingested into the Generative AI Model and the produced Datasets and from there, try to ascertain what is actually “fake” amongst the Datasets. This is an iterative process which keeps on cycling through the GAN so that eventually the reproduced or “fake” Datasets will look like the real Datasets.
  • The Training: This is the actual iterative process just described. But, once this process is deemed to have been completed, the permutations that have been fed into the “Generator” are further modified, based on the number of cycles that are needed for the “fake” Datasets to closely mirror and correlate with the real Datasets. While this can possibly be done on an automated basis, it is strongly advised human intervention be applied at this point.
  • The Equilibrium: It should also be noted that that as the iterative process comes to a point of completion, this is technically known as the “State of Equilibrium.” This simply means that it is, from the standpoint of statistics, for the Discriminator to distinguish what is the “fake” and real data, even on a granular level. Further, it is at this point, that how and where the permutations should be modified needs to be taken into serious consideration.

Some of the typical applications for using a specific type of GAN include combining and superimposing various and different images into one main, composite image, and further augmenting datasets, when and where as needed, if not enough real datasets can be procured.

The use of GANs in Computer Vision

GANs are also heavily used in Computer Vision. This is yet another subfield of AI in which a model tries to replicate the vision process of a human being. Although this is still far from being a reality, Computer Vision is quite useful in breaking down the pixels of a particular image into its most granular level of detail for further analysis.

GANs are also used in Computer Vision when it comes to CCTV technology. In these cases, the camera also makes use of Facial Recognition technology to help positively identify any individual of interest that has been recorded by the camera. An illustration of Computer Vision can be seen below:

Source: Shutterstock)

Up Next: The Diffusion Model

The next article in this series will address the Diffusion Model, a recent advancement in AI that draws upon the concepts of Quantum Mechanics and Computer Science.

 

Join the conversation.

Keesing Technologies

Keesing Platform forms part of Keesing Technologies
The global market leader in banknote and ID document verification

+ posts

Ravi Das is a Cybersecurity Consultant and Business Development Specialist. He also does Cybersecurity Consulting through his private practice, RaviDas Tech, Inc. He also possesses the Certified in Cybersecurity (CC) cert from the ISC2.
Visit his website at mltechnologies.io

Previous articleRwanda’s New 2,000 RWF and 5,000 RWF Banknotes
Next articleEES and ETIAS: the EU’s New Border Management Systems [UPDATE]