Deep Learning has seen some massive investment from some of the tech industry’s biggest names in recent years - including Google, Microsoft and Facebook - and the innovations being seen are having a massive impact across the board.
Deep learning, put simply, uses algorithms to train systems. It builds and trains neural networks by using a set of generative, hierarchical learning mechanisms to autonomously generate high-level representations from raw and unsupervised data sources. It then uses these representations to carry out typical machine learning tasks, such as classification and clustering.
One area that it has particular implications for is image recognition, and much investment has been made to develop it so as to improve technology in the field. The importance of image recognition to a firm like Facebook is clear, as it can greatly improve the website’s functionality. Another avenue that we are seeing it have a substantial impact on is advertising.
The ability to recognize a product and then show either where to buy it, or to bring up other pictures of similar products that the user might like, is a clear boon to advertisers. For example, in a picture of a movie star, Deep Learning programs can identify items of clothing that they are wearing and then pull up images of related clothing. It can also do things such as recognize a holiday destination, then, having done so, bring up images of the same holiday destination.
Pinterest is one firm leading the way in using Deep Learning for visual search. With its massive store of images, it is in a prime position to do so. Its task is also made easier by the wealth of text it has available, through both pin descriptions and board titles, that people post along with their images and can act as a starting point from which Pinterest can search, reducing the amount of computing needed.
Using the vast amount of data available to them in conjunction with Deep Learning algorithms, they have developed a new technology called visual search. Visual search uses Deep Learning to find and display images, shown to users as Related Pins. These are not based on popularity, and don’t need text labels, so can include brand new images. Their aim is to really understand what’s in an image, so as to give people exactly what they want. This aim sits in line with the aim of the modern digital marketer - to make the user feel as if they are getting as personalized experience as possible. It provides massive scope for greater targeting.