Exact sampling of determinantal point processes with sublinear time preprocessing

We study the complexity of sampling from a distribution over all index subsets of the set with the probability of a subset proportional to the determinant of the submatrix of some p.s.d. matrix , where corresponds to the entries of indexed by . Known as a determinantal point process, this distribution is used in machine learning to induce diversity in subset selection. In practice, we often wish to sample multiple subsets with small expected size from a very large matrix , so it is important to minimize the preprocessing cost of the procedure (performed once) as well as the sampling cost (performed repeatedly). For this purpose, we propose a new algorithm which, given access to , samples exactly from a determinantal point process while satisfying the following two properties: (1) its preprocessing cost is , i.e., sublinear in the size of , and (2) its sampling cost is , i.e., independent of the size of . Prior to our results, state-of-the-art exact samplers required preprocessing time and sampling time linear in or dependent on the spectral properties of . We also give a reduction which allows using our algorithm for exact sampling from cardinality constrained determinantal point processes with time preprocessing.
View on arXiv