Determinantal Point Processes in Machine Learning

Determinantal Point Processes (DPPs) are elegant probabilistic models of
diversity over discrete sets of data items. These processes enjoy
increasing interest in machine learning, and have found applications
ranging from summarization and pose estimation to computational biology.
One of their distinguishing features is the computational tractability
of basic tasks such as computing partition functions, sampling, and
extracting marginals. But despite being polynomial-time, these tasks
remain infeasible for large data sets. In this talk, I will give an
introduction to DPPs, and outline ideas for accelerating DPP sampling
for machine learning.

This is joint work with Chengtao Li and Suvrit Sra.