This project considered mathematical and computational approaches to fundamental problems in algorithmic game theory and in data privacy, with a particular emphasis on techniques form computational learning theory. Results of this research include: - Analysis demonstrating that individuals in some competitive settings may actually improve outcomes for everyone by playing selfishly rather than by following any Nash equilibrium. This is somewhat surprising, and relies on an proof that natural rules for choosing actions in repeated interaction settings might result in behaviors that cycle and never converge. - New protocols that users might follow in communication settings where they share a communication channel, in order to determine who gets to transmit when. These protocols are optimal in a certain sense. - New models for how individuals experience harm from privacy losses. These models also enable new algorithms for eliciting participation in surveys on sensitive topics. - New algorithms for privacy-preserving data analyses. This work has also demonstrated the practical potential of a particular type of provable privacy guarantee. - New models of information-sharing in social network settings. This has allowed for new connections to classic concepts in graph theory. - New results on behavior of agents in market competition settings, helping move this style of analysis away from assumption that traders will act short-sightedly.
