Seminar Series: Yi Yu (1--2 PM)

Image
Yi Yu
November 4, 2021
1:00PM - 2:00PM
Location
VIrtual

Date Range
Add to Calendar 2021-11-04 13:00:00 2021-11-04 14:00:00 Seminar Series: Yi Yu (1--2 PM) This seminar will start at an early time due to time zone difference Meeting Link Title Optimal partition recovery: from chain graphs to lattices then general graphs Speaker Yi Yu, University of Warwick, UK, Department of Statistics Abstract In change point localisation problems, one seeks estimators of change points in chain graphs with piecewise-constant means.  We will start with presenting the optimal results in such problems, then move on discussing the rectangle partitioning problems in d-dimensional square lattice graphs.  Due to the increase in dimensionality, the successful L_0 penalisation methods become NP-hard.  We will discuss how one can still achieve optimality in lattice graphs with computational-efficient methods.  Finally, we move on to general graphs, which are solely characterised by the effective-resistance connectivity, and complete the full story. VIrtual Department of Statistics webmaster@stat.osu.edu America/New_York public
Description

This seminar will start at an early time due to time zone difference

Meeting Link

Title

Optimal partition recovery: from chain graphs to lattices then general graphs

Speaker

Yi Yu, University of Warwick, UK, Department of Statistics

Abstract

In change point localisation problems, one seeks estimators of change points in chain graphs with piecewise-constant means.  We will start with presenting the optimal results in such problems, then move on discussing the rectangle partitioning problems in d-dimensional square lattice graphs.  Due to the increase in dimensionality, the successful L_0 penalisation methods become NP-hard.  We will discuss how one can still achieve optimality in lattice graphs with computational-efficient methods.  Finally, we move on to general graphs, which are solely characterised by the effective-resistance connectivity, and complete the full story.