Skip to main content
Tech Talk machine learning
Inside WEX

Machine Learning: A WEX Tech Talk

September 28, 2017

Machine learning has been around since the 1950’s when Arthur Samuelsson, an IBM researcher, created an algorithm that learned how to play checkers. After playing itself thousands of times to hone its skill, the algorithm became good enough to beat the Vermont state checkers champion. Those humble beginnings set the stage for a recent WEX Tech Talk on machine learning by ClearView Solutions Architect Daniel VanderMeer.

“These early algorithms showed a lot of promise and people really thought they’d take off in the next 10 to 20 years, but that didn’t happen because computers weren’t yet powerful enough to use those algorithms to their full effect. This AI winter started to thaw in the ‘80’s, and today we’re seeing an a explosion in machine learning applications that are revolutionizing a broad range of industries, including payments,” said VanderMeer.

The Power of Machine Learning

VanderMeer is passionate about the possibilities of machine learning, and admits that when he first started looking into it, “it looked very strange, but the underlying algorithms are pretty simple.” He shared with Tech Talk attendees some examples of machine learning that illustrate both the power and potential of this technology:

Self-Driving Cars: Not a thing of the future anymore, self-driving cars run almost exclusively on machine learning, and VanderMeer predicts that, despite possible legal hurdles, they’ll be on our streets sooner than we think.

Image Recognition: Facebook’s ability to automatically recognize and tag your friends in a photo you post and the depositing of a check using a photo taken by your smartphone are everyday examples of machine learning that matches the human capability to recognize objects in images.

Speech Recognition, Synthesis and Translation: Machine learning algorithms embedded in commercial products are the driving force behind digital assistants like Siri and Amazon’s Alexa. There’s even an app that lets you point your smartphone at a menu or sign in a language you don’t understand to get a translation.

Understanding and Decoding Human DNA: Perhaps the most dramatic example provided by VanderMeer are machine learning algorithms that decode DNA using blood samples. These algorithms can sort through the incredibly complex information stored in our DNA to predict height, weight, eye color and even create a remarkably accurate facial image.

These machine learning advances and others are possible mainly because of the confluence of three factors that will continue to drive the development of even more sophisticated products, services and applications:
• Computing power that is now fast enough to run algorithms at a rate that supports more advanced machine learning
• The availability of ubiquitous and accessible data – drawn from everything from Fitbits to credit cards – used to train algorithms to do incredible things
• Algorithm sophistication has sparked a revolution, most notably with the introduction of the deep learning algorithm that is the basis for most major products using machine learning today

Machine Learning at Work at WEX

At WEX, we deploy the power of machine learning across a broad range of our products and services. VanderMeer cited our ClearView Advanced analytics platform, which analyzes business fleet purchases, as a prime example. Developed to uncover fraud, manage product type controls, and identify cost-saving opportunities, the ClearView algorithm is at the heart of solving a simple, but significant, problem for our customers and us.

“Our fleet customers identified product miscodes sent to us by stations as their most important problem, causing noise in the data that can make it hard to get an accurate picture of purchases. Even though merchant miscoding impacts less than two percent of our transactions, it can trigger false positives for our customers and cloud the accuracy of our price indexes,” noted VanderMeer.

Our solution was to build the ClearView algorithm that has learned from lots of data from good sites, giving it the ability to detect and cleanse miscodes from the small percentage of sites sending bad data. Customers using ClearView can opt to view their data in both its original state and following adjustment by the ClearView algorithm, showing where modifications were made. Just like you’d like to get rid of junk mail, ClearView eliminates the noise and the headache of unreliable data.

VanderMeer closed out his Tech Talk with a nod to WEX data scientists throughout the organization who are exploring more uses of machine learning, as well as enhancing its current use in many of our existing offerings.

Stay connected

Subscribe to our Inside WEX blog and follow us on social media for the insider view on everything WEX, from payments innovation to what it means to be a WEXer.

"*" indicates required fields

Find out how WEX can help grow your business