Information Theory with Kernel Methods

Explore the intersection of information theory and kernel methods in this 59-minute conference talk from GSI. Dive into the challenges of estimating and computing entropies of probability distributions in data science. Examine situations where distributions are known only through feature vector expectations, focusing on the case of rank-one positive definite matrices. Discover how covariance matrices can be utilized with information divergences from quantum information theory to establish connections with classical Shannon entropies. Gain insights into advanced computational techniques for handling complex probability distributions in various data science applications.