“Deep learning” will get you 20.9 million hits on Google, 2.3 million hits on Google News, and 380,000 hits on Bing. You won’t want for information if you research deep learning, but you may still want for explanation.
That’s because deep learning, like a lot of computing concepts, is complicated. You’ll see a lot of similar descriptions of what deep learning is, and what it does, but those descriptions only suggest how deep learning works. In this post, I’ll give you a better explanation of what deep learning is, what it has already done, and its potential for business intelligence.
What is deep learning?
The average definition of deep learning goes something like this: Deep learning is an advanced type of machine learning that “imitates the workings of the human brain in processing data and creating patterns for use in decision making.”
I like to think I’m pretty smart, but I realized I had no idea how the human brain processes data or creates patterns.
What does it actually mean that deep learning works like the brain? That it gets distracted by song lyrics you can’t place? That it remembers funny things at inordinately inopportune moments? That it would rather spend time and energy recalling a third-grade little league game than the need to pay the utilities?
The whole deep-learning-is-like-your-brain comparison means this: Your brain is a pattern-identifying, category-creating machine. So is deep learning software. Both the brain, and deep learning software, use those categories to think (or compute) in a more effective and efficient fashion.
While it may seem cute when a toddler struggles with categories (that’s not an Irish Setter, little Reginald, that’s your Aunt Tiffany), the process of being able to lump separate pieces of data into categories is actually a vital accomplishment. Learning to group all those red-haired, four-legged animals together makes higher-level thinking easier. It’s also one of the most characteristically human things the human brain does.
Learning these categories paves the way for higher-level thinking. Instead of having to analyze every new red dog, the brain recognizes that dog as part of a group: Irish Setters.
Like the human brain, deep learning software is great at recognizing patterns, and creating categories from them. In computer lingo, these categories are called intermediate representations.
Antoine Amann, founder and CEO of social media automation company Echobox, explains how intermediate representations make it easier to think:
“Someone who doesn’t know anything about art is given a famous painting (say one of Van Gogh’s sunflower paintings) and asked who painted this artwork. They will have to go through a catalog of artworks until they happen upon the page that shows the painting. Someone who knows a little bit about art might know that the sunflowers are a 19th-century European artwork, so they can go straight to the relevant section of the catalog. In this case, the abstract idea of a ’19th century European work of art’ is an intermediate representation of the Van Gogh painting (and all the other paintings in that section of the book). Of course, someone who can use intermediate representation will be much faster.”
Deep learning algorithms, like the ones used by Echobox, create intermediate representations. Doing so allows their software to process data sources faster, and more accurately. One well-publicized, recent example was a Google deep learning program that was able to look at millions of pictures, and recognize which ones are cats.
Why does this matter to businesses? Because deep learning software can recognize more than adorable cat photos. It can also recognize fraud, churn rate, and future demand for a product. In many cases, it can recognize these things better, and faster, than humans can.
Deep learning success stories
Echobox helped a medium-sized celebrity magazine recognize which social media strategies would be successful. The magazine used Echobox’s automated testing feature to determine what sort of headlines were most effective. They realized “that withholding a celebrity’s name from the headline creates more clicks, but also more negative feedback,” said Echobox CEO Amann.
As a result, the company adapted their social media strategy and “settled for sharing only a fraction of posts in a click-baity manner,” and then, only when they determined it wouldn’t alienate readers.
This is all to say that a software program learned to recognize what alienates people more effectively than pollsters, or most entertainment experts.
Deep learning also helped a military aerospace manufacturing plant save energy and automate their processes. AI consultant Pavel Romashkin of Volitant AI explained how his company’s deep learning software reduced energy waste. By analyzing all the plant’s assets with deep learning, Volitant determined which machines were “creating bottlenecks in the energy flow, thus reducing energy efficiency.”
The plant was able to use that data to figure out which units needed to be repaired, which led to an 18% improvement in energy efficiency. Energy isn’t the only thing that deep learning saved, however.
Having fixed several energy efficiency problems, Volitant next helped improve future performance. Their deep learning solution was trained to manage the plant’s assets, according to how much energy each would need.
“For example,” Romishkin explained, their deep learning software “has learned how to plan to stop or reduce the load on boilers, compressors, and turbo generators, according to the demand.”
Volitant’s deep learning software also managed to fix a problem for many managers: getting the AC level right. Moreover, it determined how to “make sure the comfort standards are achieved with the lowest cost.
Deep learning is still advanced enough that not many SMBs have caught on, but as the estimated 2,000 deep learning vendors in the space expand, they’ll no doubt develop more options for small and medium businesses. The same things that larger businesses use deep learning for such as demand prediction, customer churn, and a customer’s likeliness to buy a recommended product, are all things that can benefit SMBs, too.
Deep learning’s potential to improve business outcomes
Business intelligence software vendors are also recognizing the value of deep learning. In May 2017, BI giant Tibco purchased Statistica, a data science platform. Tibco noted that the acquisition will “provide rigorous modeling and validation tools for machine learning and deep learning.”
Gartner research (full content available to Gartner clients) corroborates deep learning’s value, noting that it will be useful for everything from diagnosing diseases to determining the likelihood a customer will take a product recommendation. Deep learning will be particularly good at combining disparate types of data—especially sources “distributed over time and space.”
While some companies are building their own deep learning algorithms and applications, buyers will most likely use deep learning as part of third-party services (like Volitant and Echobox), rather than create their own (at least at this stage of deep learning’s development). Packaged applications, or deep learning APIs, will be how most companies experience deep learning.
If you’re interested in deep learning’s ability to help you keep customers and predict what they’ll want, check out vendors like Microsoft Azure, Intel’s Nervana Cloud, or Amazon’s deep learning platform on AWS.
More deep learning resources
Has your business used deep learning software? If so, I’d love to read about it. Let me know in the comments below!
If you’re interested in learning more about deep learning and machine learning, check out one of these Capterra articles:
Looking for Business Intelligence software? Check out Capterra's list of the best Business Intelligence software solutions.