It is likely that no single technology will affect your business (and probably your life) more than deep learning. It’s an idea that is achieving some public notoriety, since it enables many revolutionary capabilities – and not least self-driving cars. Deep learning is a set of techniques that derive from neural networks. This was the most promising candidate for building real artificial intelligence in the 1980s, but because of limited computing power and some techniques that didn’t scale well, the whole thing went flat, and almost disappeared. Then in the early part of this century a few researchers reinvigorated neural networks with some novel techniques, and of course computers had become much more powerful. These neural networks on steroids are collectively known as deep learning.
The application of deep learning is very broad, including image recognition, object recognition (within images), speech recognition, time series forecasting and traditional classification tasks. This in turn leads to a very broad range of business applications. Obvious ones include chatbots (automated help desk agents for example), document processing (identifying topics in text, sentiment analysis etc), optimized digital advertising, biometrics, supply chain forecasts, sales forecasts, fraud detection – and so on.
So why the term deep learning for deep business? Well, the bad news is that deep learning does many things better than humans – much better. The technology has developed to the point where deep learning powered object recognition is superior to that of humans – and visual processing is one of our top skills. It means that deep learning will correctly identify an object in an image that may be missed or misinterpreted by us. This has immediate application in the medical field, with deep learning based interpretation of CT and MRI scans already being implemented. Another application is scanning items on a production line for faults.
While image and language processing are at the sexy end of deep learning, we shouldn’t forget that it can perform very complex classification tasks much better than most existing methods. Identifying fraudulent activity, selecting the best customers to target with a promotion, picking the prospects most likely to close, and many other classification tasks are all made possible by deep learning. Of course the users of the technology may not even suspect that deep learning is powering their apps, but increasingly it will be.
The technology is moving at a frightening pace, with specialized hardware and services being developed to provide the compute power and software needed to make it all work. IBM is the most prominent technology supplier with its Watson platform, and has chosen the term cognitive computing to spearhead its move into this domain. Obviously Google, Facebook and others are developing the technology for their own uses, and particularly in the development of new products (driverless cars and object recognition in images for example).
The pace of progress will ramp up as deep learning is applied to Internet of Things (IoT) data, allowing complex patterns to be processed in near real-time. Applications in manufacturing, capital markets, retail and transport are obvious.
Finally it needs to be said that deep learning is quite different from most other attempts at AI. Andrew Ng, a leading figure in all of this, put it quite well. Prior to deep learning, everything else was curve fitting (modifying parameters until a method worked). Deep learning emulates the way we learn. Supply an object recognition network with thousands of images and it will learn what a cat looks like without being told – just as a child would. Once we have told a child that the object is a cat, it applies that label to all cats – as deep learning does. Many businesses are already starting to use the technology. One retailer of bathrooms is using deep learning to match customer profiles with the images of bathrooms, to find which styles are most likely to appeal to various categories of customer. Bear this in mind next time you take a shower in your new bathroom.