Deep learning, a new and growing area of machine learning, is widely regarded as an important step forward on the path toward true artificial intelligence. Tech firms such as Facebook and Google, as well as companies like Bloomberg (which focuses on financial technology and information), are already beginning to incorporate the technology into their product development. And while the exact definition of deep learning is still somewhat fluid, opportunities are growing for experts who can apply the technology to everything from speech recognition to bioinformatics, drug discovery and equities trading. The global analytics firm SAS says deep learning trains computers to perform human-like tasks such as recognizing speech, identifying images or making predictions. “Deep learning can break patterns into sub-components and build models very accurately,” Wayne Thompson, the chief data scientist of SAS Data Science Technologies in Cary, N.C., explained. “For example, it learns the sub-components of images to learn, say, facial recognition. And it does it with a 6 percent error rate, which is less than the human error rate would be.” Data scientists consider deep learning to be “an evolution” of machine learning, added Gideon Mann, head of Data Science for New York-based Bloomberg L.P. Mann describes it as “the branch dedicated to huge data sets.” While its use is something of a niche in software engineering, he added: “If you’re in data science, you need to know about it.”

Deep Learning in Context

In discussing deep learning and machine learning, data scientists make it clear that the two are intrinsically linked. At least in the immediate term, Mann believes it’s more important for software engineers to understand the basics of machine learning. “Software engineers would benefit from learning how to phrase, and set machine learning problems, to understand the lifecycle,” he said. “That way, they’ll know how to interact with data specialists and understand the context of their work.” Right now, he believes, “It’s more important to broadly understand machine learning and the problems you can solve with it.” At the same time, Thompson thinks engineers who are “application builders” should have a grasp of both deep and machine learning. He sees a time in the not-too-distant future when developers will create “modules” using deep learning to look at images or video of computer chips to detect flaws, or scan x-rays to determine the spread of a disease and help physicians plan treatment. Deep learning allows users to plunge much deeper into their data, Thompson said. “You can take 20 factors of something like blood pressure and break it into 10,000 factors,” giving data scientists, analysts and other users a more refined look at the information in front of them.

Putting Deep Learning to Use

While Google and Facebook have been very public in how they’re using deep learning—Google relies on it to match ads with search queries, while Facebook applies it to tailor newsfeeds to each user’s individual interests—both Mann and Thompson see machine and deep learning as useful to a variety of industries beyond tech. For example, Mann said Bloomberg finds machine learning “increasingly indispensable” as it develops products to help financial companies operate in markets that are faster, global and more complex. He predicts insurance companies will harness deep learning to more accurately inventory insured assets, among other things. Automakers—and not just Tesla—are showing “a huge amount of interest” in it as they develop self-driving cars. “You can’t overstate the impact deep learning will have on product management,” Mann said. Product managers, he believes, will “need to understand the underlying (deep learning) task in order to understand their product.” With deep learning, a product can essentially train data in ways that can create new offerings. For his part, Thompson thinks the technology is a critical component of cognitive computing. “Deep learning is built into the natural language interface,” he said. “You marry what the user says with the natural language interface and deep learning to get an answer.”

Where to Begin

If you’re a tech pro who wants to learn more about deep learning, where can you start? A number of online courses are available from sites such as Coursera, edX and Udacity, “which have well-sized lectures that are built out,” Mann said. Of course, professional journals offer much information, although that can prove an unnecessarily difficult way to get grounded in the basics. Thompson thinks that classes like those offered by SAS are also valuable, and suggests you check out KDNuggets.com, which is something of a hub for many data scientists. Whichever avenue you choose, many machine learning courses will include material about deep learning, as well. And remember, Thompson adds, that you’ll want to learn the languages used to take advantage of deep learning, such as SAS, Python, SQL and Hadoop. “It doesn’t matter what kind of data you throw at it,” Thompson observed. “Anywhere you have lots of data, deep learning is the latest proven mechanism of data analysis.”