|
Mathematicians, statisticians and computer scientists have developed
a new mathematical technique for identifying high-risk drivers. It's
a breakthrough that could save both the insurance industry and its customers
millions of dollars each year.
Who says mathematics isn't relevant? It is for people with rising
insurance rates. A team of researchers at Université de Montréal has devised
a new statistical tool that could lead to lower insurance premiums for
good drivers and higher profits for car insurance companies.
|
Helping Computers Learn
Can machines learn? Apparently so.
A growing number of professors and students in Canada are specializing in statistical machine learning - an emerging discipline in which statisticians and computer scientists collaborate to develop techniques that enable computers to "learn" useful information from large amounts of data that contain numerous variables.
"It's about using different techniques for extracting useful information from data," says Dr. Yoshua Bengio, a MITACS researcher and Canada Research Chair in statistical learning algorithms at Université de Montréal.
In one MITACS-supported project, Dr. Bengio and his collaborators are developing statistical models that could help pharmaceutical companies reduce their drug development costs by better predicting which chemical compounds are potentially most likely to help a particular medical problem.
Adds Dr. Bengio: "It helps to have an organization like MITACS supporting research and students in an area where Canada ranks very strong internationally."
|
|
What began as a MITACS-sponsored research project has evolved into a new commercial product and a university spin-off that is helping insurers analyze copious amounts of data to more accurately predict who might have a car accident. For the majority of drivers who are considered low-risk, the payoff could be lower premiums.
Current statistical models have been difficult to apply to the insurance industry, which contains so much data and so many variables that it's difficult for statisticians to develop probability models that accurately predict how risky a customer may be."The methods currently used in car insurance are not very discriminating, meaning the good drivers pay more than they should and the bad drivers pay less than they should," says Dr. Yoshua Bengio, a computer scientist at Université de Montréal and a principal investigator with MITACS. "We've come up with a way for the insurance company to decrease premiums for the less risky customers, which is the majority of people. Companies could then use this as a marketing tool to increase their market share and their profits."
Dr. Bengio is also the Chief Scientific Officer and one of four founders of Apstat Technologies Inc., a Université de Montréal spin-off that last year began marketing this new data-mining product to car insurance companies. The product stems from more than a decade of work by Dr. Bengio and his students.
"Our initial research was done with two Networks of Centres of Excellence, MITAC and IRIS (Institute for Robotics and Intelligent Systems)" he says. "We later partnered with a large North American automobile insurer to transfer the technology. This is what prompted three of my students to start a company to commercialize similar techniques." Those students turned entrepreneurs are Charles Dugas (Chief Executive Officer), Nicolas Chapados (Executive Vice President) and Pascal Vincent (Chief Technology Officer).
The market potential for Apstat is huge. As Dr. Bengio points out, the insurance industry has been using almost the same mathematical techniques for decades.
Meanwhile back at the university, Dr. Bengio is working with a growing number of researchers, students and industrial partners to develop similar data-mining solutions for other applications, including drug discovery, statistical language modeling, and telecommunications marketing.
"This is a very rapidly growing field in universities across Canada. I credit organizations like MITACS not for only supporting this basic research, but also for encouraging us to find partners and look for ways to transfer our technology," he adds. "There's still this perception that mathematics can't be related to useful things, but that's wrong. Some of the things I do involve very complicated mathematics that lead to very important applications."
www.mitacs.ca
|
