Digital Marketing with Generative Adversarial Networks

A new technique called Generative Adversarial Networks, or GANs, has been showing promising results in the field of artificial intelligence and computer vision. GANs are being used to create photorealistic images, detect pedestrians in complex and noisy environments, and even generate music! In this article, we’ll explore how to use GANs to improve your digital marketing strategies using neural networks and adversarial training...

What are GANs?

Goodfellow et al. published their paper introducing a new family of models: generative adversarial networks (GANs). These were originally devised to improve unsupervised learning for image classification problems, but have since found many more applications in computer vision and natural language processing.

A GAN consists of two neural networks: a generator and a discriminator. The generator takes random noise as input and outputs synthetic data. The discriminator receives both real examples from your dataset and examples from your generator network, it is responsible for classifying whether an example is real or fake. In theory, these two networks will learn to balance each other out; as one gets better at creating realistic images, it forces its opponent to become better at distinguishing between generated and real data so that they cannot be easily fooled by unrealistic-looking examples.

How are they useful in digital marketing?

Because of their ability to synthesize high-quality content from human inputs, GANs are useful for training algorithms that can learn new content on their own. Unlike most neural networks, GANs have two competing outputs: one which tries to fool a second network into thinking it is real data, and another which tries to distinguish between fake and real data. This competition makes GANs incredibly effective at generating realistic images that fool humans — all of which is critical in digital marketing. If a customer clicks an ad or shares your website, you need your website’s copy to be real enough for them not to notice it’s generated automatically. If you want to keep customers engaged after they click through, you also need them not to notice any differences once they land on your page. The longer someone stays on your site, whether manually or by following an advertisement or sponsored link, the more likely they are to become a long-term customer. By teaching computers, how to generate realistic copies of text and images, we will make ads more appealing by producing higher quality advertisements—ones that don’t jump out as obviously ads. More compelling ads mean higher click rates (through fewer perceived fakes) and ultimately greater ROI when we funnel traffic to our websites.

Put simply, generative adversarial networks could be key to driving down advertising costs and convincing users to engage with brands even if they know what they’re doing is advertising. But there’s no need to wait until these technologies are developed; marketers already use creative design workarounds such as image blurring, background removal, duplicating words from existing images, etc. These techniques can take months or years of professional expertise to master, but a GAN can effectively do them within minutes—which means we will get better results faster using GANs than hiring humans.