Event type On-site Event
LocationB1.1.02 NOI Techpark BZ
Departments ENG Faculty
Contact Prof. Giuseppe Di Fatta
Giuseppe.DiFatta@unibz.it
Accelerated Computation and Minimax Lower Bound of Kernel Stein Discrepancy
Zoltán Szabó is a Professor of Data Science at the London School of Economics. His research interest is statistical machine learning with focus on kernel methods and their applications.
Event type On-site Event
LocationB1.1.02 NOI Techpark BZ
Departments ENG Faculty
Contact Prof. Giuseppe Di Fatta
Giuseppe.DiFatta@unibz.it
Kernel Stein discrepancy (KSD) is a powerful tool to quantify goodness-of-fit on a wide variety of domains with numerous successful applications. However, the classical KSD estimators (relying on U- and V-statistic) scale quadratically in terms of the sample size, which hinders their application in large-scale settings.
In this presentation we will
- present an accelerated KSD estimator based on the Nyström technique while preserving the statistical accuracy of the quadratic-time KSD approximations,
- and settle the optimal rate at which KSD can be estimated.