Brendan Frey, a professor in the department of Electrical and Computer Engineering at University of Toronto, published a paper on what he called “affinity propagation” in the February issue of the academic journal Science. Affinity propagation allows developers to create software that measures the similarity between a set of data points. It does this by exchanging messages between many pieces of data all at once, rather than trolling through data sequentially to choose between one piece over others.A database that tries to recommend movie titles based on the selections of other users, for example, could work more quickly through affinity propagation, Frey said. In a traditional approach, software would work with a random sample of movie selections and winnow them down. Machine learning software based on affinity propagation would send messages back and forth about all the movie selections and isolate those that matched in some way, sort of like the way brokers in a stock exchange hustle and bustle their way to make a deal on a trading floor, he said.
The algorithm could be integrated into applications in business intelligence, customer relationship management and enterprise resource planning. “It has very broad applicability,” Frey said.
Frey’s team first developed a customized version of the algorithm to help biology researchers who were sorting through some 75,000 DNA segments.
The algorithm and some software Frey and his team developed has been posted online, but Frey said he hasn’t determined the commercial path for affinity propagation. “As a professor, my responsibility is primarily to society,” he said.