What's unique about the work being done by IBM is the scope and scale to which the mathematical modeling techniques are being applied, "vs. the techniques themselves," Morency explains. "The military has certainly been using these types of techniques to support battlefield and rescue strategies for a while. The IBM work is the first effort that I am aware of to apply these techniques to large-scale, non-military disaster simulation for the purposes of reducing logistical complexity and improving overall emergency response."

According to Schieber, predictive analysis involves making predictions about future events based on historical and current data. He sees stochastic optimization as the next step: "Given the prediction about future events and some probability distribution associated with these events, find an optimal policy... [In] the game "21" the 'predictive analysis' part would be to predict the probability of the next cards based on the cards that are out already, and the 'stochastic optimization' part would be to decide what is the right action -- stop or go -- based on these predictions."

From recalculating limos to transforming healthcare

When concern about pandemics began surfacing a few years back, IBM wanted to examine the impact an outbreak would have on its own infrastructure and its ability to serve clients. "For example, if there's a decision to close airports, to stop all air traffic when such a pandemic occurs, how would this impact the spread of the pandemic?" Schieber explains. "Here we worked with a team of doctors from places like Johns Hopkins and government agencies to try to pick their brains and understand what would be the change in probability in infection if we asked everyone to stay home or forced everyone to go with a mask."

The scenario addressed by the Stochastic Theory can also be more modest. For example, members of Schieber's team are currently working with two IBM clients on how to address the problem of shipping containers coming to the United States from China filled with goods but going back empty.

Another IBM project examined the question of how to manage scheduling for a limousine business. The company had hired IBM initially to help install GPS equipment in their vehicles and integrate it with their other computer systems. "They thought by putting GPS in their vehicles everything would be solved," recalls Schieber. The consultants' response: "You are missing the extra link -- taking this data you're getting from the GPS and putting it together into an optimized plan." The consultants brought in IBM's maths scientists to figure out what that plan might look like.

As part of their research, IBM's scientists needed to understand the nature of untimely limousine service. Using a snowstorm scenario, IBM examined whether it would be better for a limousine service to plan on being 20 minutes late in picking up two customers or on time for one and 40 minutes late for the other? "Which one is better?" asks Schieber. "We spent two days on that problem."

Typically, that sort of scenario is handled intuitively by the driver or dispatcher. But when a machine is making the decisions, the math model needs to quantify the relative merits of one type of customer over another, Schieber says.

Although these kinds of puzzles keep Schieber his mathematician co-workers busy, they also enjoy tackling problems that seem truly unsolvable -- such as healthcare. "If you look at the healthcare system, we all agree it's not efficient," he says. "From knowing this to getting something going is a big distance." Applying the Stochastic Optimization Model in that type of project will require several hospitals, he predicts, working together to define how their resources can be better applied. "Here, we see that we can do something that is not only good for the bottom line as a company but also good for society."

Dian Schaffhauser is a writer who covers technology and business for a number of publications. Contact her at dian@dischaffhauser.com.