Research at Rowan University may help rapidly advancing technology keep pace with shifting data patterns in today’s complex world.
Backed by a $164,923 National Science Foundation research grant funded by the American Recovery and Reinvestment Act of 2009, Dr. Robi Polikar, Professor and Chair of Electrical and Computer Engineering, and his team are studying new ways to help computers learn incrementally from unbalanced data in a nonstationary environment. In other words, one type of data far outweighs others in these situations while the characteristics of the data continually change.
The research team is creating an automated algorithm allowing machines to track changing patterns when data are unbalanced, enabling machines to monitor data, adjust their analysis based on changes they recognize and make decisions based on large quantities of data.
Such applications offer a number of opportunities, such as financial fraud detection and analysis of climate data and energy demand. “Predicting our future energy need — and hence the cost of that energy — and detecting different types of financial fraud and even detecting spam mail are examples of nonstationary problems,” Polikar said. “In such cases, similar circumstances in the future may lead to different outcomes, and hence the nonstationarity.”
Their investigation is ongoing, but research team members already have presented their preliminary findings internationally. “There are increasing numbers of real-world applications that generate nonstationary data,” Polikar said. “There are very few approaches out there that can address this problem. It’s a very new field.”