Earth Mover's Distance and Kernel Statistics

The Wasserstein/Earth Mover's distance combines physical distance and probability considerations to robustly compare random variables. Kernel methods in statistics uncover nonlinear and low-dimensional structures in high-dimensional data. Can one combine the two approaches to create robust kernels which takes continuity (in time or space) of data into account?

Name of Direct Supervisor: Amir Sagiv

Position Dates: 6/1/2020 - 8/31/2020

Hours: 35 hours/week

Qualifications: Basic notions in statistics, preferably some background in probability/calculus. Familiarity with Fourier transform- a plus. This is intended to be a coding (Matlab/Python) project. However, students with a strong background in probability/analysis/statistics who want to do a theoretical project are also welcome to apply.

Eligibility: SEAS only

Please send your CV and transcripts to Amir Sagiv, as6011@columbia.edu.