Is edge computing just new branding for a type of cloud computing, or is it something truly new? Let's examine how the edge approach works, where edge makes sense, and how edge and cloud will coexist.
Machine learning: 3 considerations from LexisNexis CTO
LexisNexis Risk Solutions’ CTO Vijay Raghavan offers his take on how to tap the power of machine learning
Can machine learning benefit your organization? And if so, how?
Vijay Raghavan, executive vice president and CTO at LexisNexis Risk Solutions, explains that it’s important to first understand that there’s no single ideal use for machine learning. “Machine learning is a technique rather than an off-the-shelf tool,” he says. “There is no right time to use machine learning, other than understanding what problem you’re going solve and whether machine learning is going to help solve that problem.
“It could be revenue growth, cost reduction, time to market, risk mitigation – it all depends on what your priorities are as a business,” Raghavan continues. “I could use machine learning to build models within or around my internal data warehouse to better predict customer attrition or revenue attrition.”
The key, he says, is to make sure the technique will add value in a specific way. “You need to have modelers and statisticians who understand what techniques to use and when, and the pros and cons of those techniques. Machine learning isn’t just one thing – it’s a toolbox full of tools.”
That said, here are three factors to keep in mind when considering how your organization might benefit from machine learning.
1. Beware of biases and assumptions
“Machine learning is not limited to usage within IT,” Raghavan notes. “Machine learning techniques are tools that sit at the intersection of mathematics/statistics and computer science. As such, [machine learning] can be used by a wide variety of people across multiple fields and departments – which means it can also be misused by a wide variety of people.”
Two of the biggest mistakes: depending too much on a single machine learning approach, and not being open to what the data is telling you. “Experienced statisticians and modelers know not to fall in love with a particular technique, and [they] know how to experiment with approaches to assess the best fit,” Raghavan says. “A common mistake is to jump to conclusions because of ‘confirmation bias,’ wherein a particular technique may appear to confirm what you want to believe. This is not just a rookie mistake.”
2. Machine learning can help solve the talent crunch
The proliferation of big data, and the need for people with the skills to make sense of that data, is an ongoing challenge for most large organizations. Can machine learning help?
“Machine learning can help companies solve big data problems because it helps ease the effort of data discovery as part of the model-building process,” Raghavan explains. “This is a big part of what ML does. So to the extent that it makes modelers more productive, you could say that it helps with the perceived data scientist or modeler shortage.”
[ How will machine learning help shape future technology trends? See our related article, IT leaders weigh in on what excites them about the future of IoT. ]
How does machine learning make data modeling more productive? “Tools – including machine learning tools – that have thus far not been available to a modeler are now being democratized and made more friendly,” Raghavan says. “Previously, a data scientist had to be a polymath who understood data, statistics, modeling, computer science, big data technologies, and a few other things thrown in. That is changing as the tools become less formidable and more malleable, and the line between a data scientist and a modeler gets blurred.”
3. Consider the ethics of machine learning
Asked what final advice he’d give CIOs about machine learning, Raghavan responds, “Make sure these tools are used ethically. I saw a movie many years ago in which the protagonist tells his adversary, ‘You’re a very logical man – but to be logical is not necessarily to be right.’”
Any technology tool can be used for good or bad, he points out. “The more powerful the tool, the greater the damage it can cause – deliberately or inadvertently – especially when we are talking about statistical tools that are attempting to predict the probability of something that is or is not likely to happen.”
Finally, if you deploy something with the potential to cause damage, Raghavan adds, “Ignorance or incompetence, even without malicious intent, is not an appropriate excuse.”