Ethical trials, targeted ads
Clinical trials of new drugs or devices face a problem that most empirical inquiries don’t: They must not only provide clear data about toxicity and efficacy but also try to maximize the quality of treatment for all of the patients enrolled. Online advertisers face a similar quandary. They want to test variants of ads to see which drive more traffic, but they also want the better-performing ads to reach more viewers. In a forthcoming paper in the Annals of Statistics, Philippe Rigollet, an assistant professor of mathematics at MIT, and colleagues present a new way to organize such tests that’s mathematically guaranteed to yield the optimal result. Rigollet believes that the work could also have implications for the distribution of tasks in parallel computers. “It’s called the ‘exploration-versus-exploitation dilemma,’” Rigollet says. “It was a bit of a dormant field in the ’80s and ’90s, and then in the early 2000s came Amazon and...