Deep learning should deliver systems that not only think, but keep learning and self-directing as new data flows in. Enterprise is taking note.
Nothing gets to the heart of real-time interactions more than the constellation of technologies that come under the umbrella of artificial intelligence. And the emerging field of deep learning promises that systems will be able to not only think, but keep learning and self-directing as new data flows in. Lately, a sizable segment of organizations have been piloting or even implementing deep learning technologies to deliver self-guided systems. However, deep learning is still very much a work in progress.
That’s the word from Ben Lorica and Mike Loukides, both with O’Reilly Media, in a recent research brief that explores the growth and potential of deep learning across the enterprise landscape. They define deep learning as a part of AI and machine learning that got its start about seven years and has been captivating data scientists, computer scientists, business leaders and gamers ever since. More recently, it was a machine’s success at the game of Go (Alphabet/Google’s AlphaGo) that brought deep learning into the spotlight.
Deep learning can play a role in a range of real-time, interactive applications, including speech recognition, visual recognition, and machine translation using neural networks, Lorica and Loukides explain. However, “our understanding of deep learning systems remains a work in progress,” they caution. “As successful as deep learning has been, our level of understanding of why it works so well is still lacking.”
Deep learning survey results are in
In a recent survey of 3,300 technology professionals conducted by O’Reilly, 28% report they are already working with deep learning solutions in some capacity, they observe. Looking to the future, 54% of respondents predict that deep learning will play a large or essential role in future deep earning efforts, and another 38% anticipate using it in the future.
Most respondents (73%) indicated that they have begun playing with deep learning software. The applications of greatest interest at this time include computer vision (13%), text mining (11%), and enhancing data analytics capabilities (9%).
The main challenges to deep learning track closely to the challenges with other types of technology products — it’s hard to find (or train) people with the skills to build, implement or run these systems. Skills shortages are the leading challenge, cited by 20% of technology leaders. Finding enough compute resources follows at 9%, along with data-related challenges (8%).
The skills situation may only get tighter, Lorica and Loukides warn, noting that only 11% indicated they have hired specifically for deep learning applications. “This number seems low, but it probably reflects where companies are in the hiring cycle. If only 28% of respondents are currently using deep learning, but 54% think that it will play a large or essential role in the future, it’s likely that many companies just haven’t started hiring yet. In turn, if most companies are thinking about AI projects but haven’t started yet, the talent shortage will get much worse before it gets better.”
TensorFlow leads the list of deep learning frameworks or tools now being used, the choice of 61% of respondents with deep learning projects underway. Another 25% are using Keras, and 20% prefer PyTorch.