Reviews: Maximizing Induced Cardinality Under a Determinantal Point Process

In that framework and once a DPP has been learned, the authors remark that finding the sample with maximum DPP likelihood (the so-called "MAP recommendation") does not lead to meaningful recommendations. The contributions are as follows. The authors introduce another utility function, the maximum induced cardinality (MIC), and explain how to approximately optimize it using submodular approximations. Algorithms to optimize the MIC criterion are compared on synthetic datasets.