peshkova - Fotolia
When it comes to the promise of deep learning technology and artificial intelligence, Ace Moghimi is "pessimistically optimistic."
"Is it the end-all, be-all? Probably not," he said. "But as we work on improvements in statistical techniques [and] compute power, all these things are coming together. We see value in it."
Moghimi is global head of innovation at Toronto-based Manulife Financial Corp. and John Hancock, its U.S. subsidiary headquartered in Boston. In that role, he helped found the company's Lab of Forward Thinking, or LOFT. It looks for ways to implement new and emerging technologies to improve Manulife's core business of selling retail investment and insurance products.
One of the biggest technologies to come along in a while is deep learning. In the past year or so, this extension of traditional machine learning has exploded in popularity in part because, as a core component of artificial intelligence, it's riding that technology's rising tide. Tech companies like Google, Facebook, Twitter and Yahoo are using deep learning to classify images, decode human speech and develop computer vision.
But more traditional enterprises have been a bit slower to find uses for deep learning technology. The technique is a good match for the most complex data problems, but most of the issues facing businesses today are much simpler than replicating human brain patterns.
Finding the right fit for deep learning
That doesn't mean deep learning has no role to play in the traditional enterprise, however. One area where Moghimi and his team have found deep learning technology to be a good fit is qualitative stock research. Every day, researchers at John Hancock go through stock reports, Securities and Exchange Commission filings and news articles to assess potential investing opportunities. In the past, this was all done manually and was subject to researchers' subjective interpretations.
But free text analysis happens to be a strong use case for deep learning. Moghimi and his team have developed learning algorithms using a tool from Indico Data Solutions Inc. that ingests all these reports and analyzes the text, looking for signals that a stock is about to take off or drop in value. The algorithms then make recommendations to the researchers, who assess them and then pass them on to investing teams, greatly compressing the time it takes to evaluate all the data.
"On a typical day, researchers can't do that on their own," Moghimi said. "The ability to train deep learning models is tremendously advantageous. It makes the process much faster and more efficient."
Moghimi has been looking for opportunities to utilize machine learning algorithms over the past year or so. He said he sees deep learning technology, and, ultimately, artificial intelligence, as progress over traditional machine learning.
Challenges to contend with
But that doesn't mean deep learning is an unqualified good. Moghimi said there are still a lot of questions around how to use the technology. From a strategic standpoint, finding the right use cases is a challenge. Also, staying on top of the latest tools and techniques can be difficult in this rapidly developing corner of technology. "The most challenging part has been working through that uncertainty," he said.
All of which are reasons why Moghimi describes himself as pessimistically optimistic when it comes to deep learning. If you read any given article on the topic, you're likely to hear somebody say something along the lines of "deep learning will disrupt every industry." For Moghimi, the technology isn't likely to be that impactful, in part because of the uncertainty.
But he said he certainly sees it as an incremental improvement over more traditional practices that could pay dividends for businesses. "We're seeing it work," he said. "There's definitely value there."
Necessary system infrastructure for deep learning remains uncertain
Big data fuels the rise of deep learning and artificial intelligence
TensorFlow aims to become the platform of choice for deep learning