Reviews: Learning Tree Structured Potential Games

The problem presented in the paper is interesting. In general the notion of using games and their results as samples is intriguing. The locally-optimal nature of equilibria seems interesting to explore, especially since many methods (such as structured prediction) usually focus on MAP assignments. While the general idea is interesting, the experiments are disappointing, and suggest that the method is computationally prohibitive and cannot be used in practice. The lack of guarantees on the optimization process is also discouraging.