Peripheral Image Customization using Machine Learning
Abstract
In the age of data overload where human recipients are targeted with advertising information ranging from email campaigns and search engine ad delivery to billboard advertising and clever product placements in their favorite sitcoms, the challenge has always been to capture the recipient’s attention. As more and more advertisements compete for the recipient’s attention in a highly targeted fashion, today’s ad recipients have a much higher threshold to perceive the ads served to them from among various sources.
Ads are most successful when they appeal to the idiosyncrasies of the targeted audience. The more personalized an ad the more likely its success. However, even amongst a carefully chosen demographic having very similar characteristics, ad recipients can have a broad range of likes and dislikes.
Serving highly specific ads to each targeted individual, without doubt, solves this problem. Such an approach, however, ignores the finite resources, especially for smaller businesses, that prohibits creating a highly targeted and specific but large number of creatives/ads to each of the targeted recipients. This research targets the principal challenge that exists for all forms of advertising: how can ads be made as personalized as possible to their targeted recipients, while at the same time acknowledging finite resources (especially human) that go into creating these ads?
By making use of Machine Learning methods such as Deep Neural Networks, Generative Adversarial Networks (GAN), Variational Autoencoders (VAE), we try to conquer this challenge.
