Deep Learning Is Saving Data Centers—Here’s How
The concept of a data center is simultaneously fascinating and frightening. A data center fascinates—enthralls even—due to what it is capable of doing. Trillions of computations performed on an hourly basis, carrying everything from the operations of a business to the various tentacles of the Internet of Things that we hold in our pockets, cars, and kitchen appliances.
At the same time, a data center is an unfathomably hungry, voracious eater of energy. Each action performed by a computer needs energy, and that energy comes mostly from fossil fuels such as coal. Artificial intelligence has the potential to swoop in and save the day, solving this problem, and others. The true potential of artificial intelligence is found in its ability to perform deep learning. An examination of what deep learning is and the issues facing data centers now quickly shows why the two are fast becoming great friends.
How Deep Learning Works
A computer, at first, was just something that was used to crunch numbers. It would have certain rules programmed into it and those rules, when given input, were able to produce sets of data. Think of a calculator, for instance. Inside the calculator, you have a long list of rules that govern how different calculations are performed. When we enter numbers and what we want the calculator to do with them, a result is spit out. However, deep learning takes the depth of computer calculation and makes it something that’s much more like “thinking.”
When a computer engages in deep learning, it is using algorithms. Algorithms are capable of taking variable sets of data and incorporating them into a formula. Because of the variable nature of what goes in and what is produced by an algorithm, the computations performed by algorithms look quite different than those performed by a calculator, and the results are more dynamic—and powerful. With deep learning, a computer can learn how to classify things in a way similar to a human. This can then be used to further enhance the learning in order to support the generation of more classifications and distinctions based on each one. Let’s look at a classic example.
Deep Learning with Vehicles
Deep learning is able to see, read, and hear. One thing a deep learning machine can do is see vehicles. A deep learning computer can see a hundred vehicles and then start to classify them based on what it sees, or their attributes. If, as it looks at the vehicles, it sees 50 cars, 30 trucks and 20 motorcycles, it is then able to classify each one according to its attributes. For instance, a truck may be slightly larger, taller and with different spaces between the wheels and chassis than a car. Therefore, the computer has taught itself what a truck is. The motorcycles would have two wheels instead of four, and the computer would thereby learn what a motorcycle is.
Deep learning could then power the process to go a level deeper and begin to classify how each of these different vehicles moves, for instance. How they accelerate at different rates, stop at different rates or have different turning radiuses. It could go a level deeper and observe the interactions between different classifications of vehicles. For example, a bus would get its own classification, and that would differ from a school bus. The behavior of other vehicles when a school bus stops would produce a new set of rules. Pretty soon, deep learning would be able to make a car drive safer and even more efficiently.
Deep Learning in the Data Center World
The applications for a data center quickly become apparent, particularly when we consider some of the problems a data center faces. As energy is used to perform various calculations, there is a problem of excessive energy consumption. However, there are some operations that are necessary to the functioning of the system. As a deep learning computer observes what is happening, it begins to classify things in terms of what’s necessary and what is not. Additionally, some of this information can be further classified into computations that can be performed at varying degrees, thus decreasing the energy load on the system. And, of course, unnecessary computations can be eliminated altogether, instantly eliminating waste.
Deep Learning Business Efficiency
If we look at business solutions as derived from a series of calculations, it is easy to see how deep learning within a data center can result in hefty dividends. Let’s think about it in terms of sales. If a product is sold over the Internet, a deep learning computer can analyze the pattern of sales according to location, the amount of sales and even the length of time it takes from when a customer views an item to when a purchase is made. It could then start making classifications of the different types of sales, the same way it could do with different types of vehicles. Perhaps it learns that sales occur in England much quicker than sales in Canada, and the resulting net revenue is greater. If it is given the ability to alter the sales system, it can allocate resources to capitalize on the purchasing patterns of people in England. Further, it could help troubleshoot the sales problems in Canada.
Granted, all of this could be done by a human or even a team of humans, but how long would it take? A few weeks? A couple of months? With deep learning within a data center that handles this process, it would only take a few seconds, if that.
The Potential of Deep Learning in the Data Center
A data center, due to its huge amount of computational resources, is the perfect home for deep learning computers. In fact, some computations would be so complex and require the processing of so much data that it wouldn’t be economically feasible for them to occur on a home- or office-based computer. This makes the data center the ideal deep learning playground. If your company learns of a deep learning system that may benefit its operation, housing the whole process in a remote data center would present a well-packaged solution.
Further, even now, deep learning is creating opportunities for efficiencies that would otherwise be impossible without deep learning. Google’s DeepMind is the perfect example. DeepMind has become the brain that powers the energy systems of Google’s data centers, delivering powerful—and compelling—results. It works by taking in all of the different data sets produced by the various energy systems of the buildings, as well as their individual components such as pumps, chillers, compressors, etc. It then is able to make recommendations for the increased effectiveness of these systems—and how to accomplish this more efficiently.
Deep learning within a data center will get to the point where a data center will be able to offer a buffet table of different deep learning modules that may help power your business. At the same time, current deep learning functionality will be helping the data center to run more efficiently, saving it and its customers crucial capital. In other words, we’ve come a long way from the good ol’ calculator.