What do you know about GANs?
In 2017, the International Foundation for Art Research (IFAR) in New York was responsible for determining that four paintings attributed to the artist Jackson Pollock were fakes. The works were in the possession of a German collector who lived in Chicago.
IFAR, a non-profit organization, has this objective: to analyze works of art and clarify their authenticity. After all, these are works that can be sold for very large sums! Jackson Pollock's paintings have even sold for 140 million dollars.
This process of separating true and false online content has taken on another dimension with GANs. In this article, we'll explain what GANs are and how you can use this technology in your work.
What are GANs
As Generative Adversarial Networks (are an artificial intelligence (AI) model that aims to discriminate and generate realistic content.
GANs work like a ring, where two boxers will face each other. One of them, the generator, will produce the content (text, image or video). The second is the discriminator, who has the ability to assess whether the content is true or false.
While the generator will look for ways to produce content that can fool the discriminator, the discriminator will analyze everything the generator produces to expose the fake content.
Because it is an AI that has the two components working in this dispute, the GANs are called adversaries. However, it's not a competition to see which one is more capable of carrying out its task successfully. The aim is for one to learn from the other in order to do its job as well as possible.
Types of GANs
As GANs are an AI model, they have a number of applications. Consequently, several GANs have been developed to perform certain tasks.
Talking about each one would make this article go on for pages and pages. So let's focus on three models and how they can be used.
DCGAN's: Deep Convolutional GAN's
Convolutional Adversarial Generative Networks were developed for images. The aim is to edit images in a convincing way, especially by removing elements that reduce sharpness.
Another important point is the generation of images that leave anyone confused: is it a real image or not?
SRGANs: Super Resolution GANs
Super Resolution Adversarial Generative Networks are a mixture of GAN models. Through this AI, the user can upload entire images with low resolution and obtain high-resolution results.
This enhancement restores the image to its sharpest characteristic.
cGANs: Conditional GANs
Conditional Adversarial Generative Networks are an optimization of other models. This optimization is designed so that the user can determine the conditions for producing texts, images or videos.
In this way, the results will be closer to what the user wants to achieve with the GANs.
How to Use GANs in Your Work
GANs have more interesting applications for professions that involve creativity. The production of texts, images and videos using AI is a way of speeding up the delivery of materials. To illustrate this immense range of possibilities, we will emphasize three options.
Ad Campaigns
Advertisements are a jigsaw puzzle that must be thoroughly tested to ensure that each item is as effective as possible. However, producing all the pieces, and their size variations, can be a time-consuming task.
So GANs are a way of speeding up the process. Content generation via AI, especially conditional models, makes building campaigns a faster task with more variables to test.
Blog articles
Another application is the production of blog articles. The biggest advantage of GANs for texts is that the generator and discriminator components will help to produce articles that are more believable.
Not only will they be more precise texts, but they have a greater ability to simulate a person's way of writing. By running the article through tools that identify the use of AI, the goal is to remain undetected.
If you're interested in this subject, read our article on Artificial Neural Networks!
Photography Services
Do you have an old photo at home? With GANs you can digitally enhance the image and get a sharper result. This is the task that many photographers around the world are receiving from their clients.
People look for this service as a way of preserving their memories of deceased parents, grandparents and relatives through their photos. It's also a way of enhancing historical documents and not losing their content.
The complicated side of GANs
While GANs are a technological innovation full of advantages, they can also be dangerous.
The ability to simulate realistic content increases the chances of malicious people producing misleading content. Fake fake newsin particular, has become a global communication problem and GANs can amplify this challenge.
Not only because of the production of false content, but also because of the discriminating component, which works to make everything even more realistic. As a result, it's to be expected that few people will be able to tell the difference between correct and incorrect content.
And think of all the viable applications of this!
Texts and articles that seek to validate false information, image editing that is practically undetectable, audio so realistic that it can easily deceive.
A current example is the use of GANs to analyze videos of influencers and presenters to simulate their voice patterns. In this way, scammers can produce very convincing audios and extort relatives, friends and fans of these personalities.
Conclusion
Now that you know about GANs and how they are used today, it's your chance to include this AI in your work.
Here on the Pareto blog you can also learn about other generative models and make your tasks faster and more effective. To learn more, just choose a topic here on the Pareto blog and follow the next posts.
Happy reading!
