Optimal transport · graphs · time series

The mathematics of moving piles of sand.

A probability distribution is a pile of sand with unit mass. Optimal transport asks how to move one pile onto another as cheaply as possible, and the answer turns out to be a geometry. This site collects research that puts that geometry to work: finding structure in time series, and averaging and decomposing distributions that live on graphs.

Looking for research or applied ML roles — available now.

Try the demo Get in touch About

A three-peaked distribution being transported onto a single-peaked one

Try it: barycentric coding on a graph

λ = (, , )

ν1
ν2
ν3

Linear mixture Σ λiνi

Mass fades in and out in place; nothing moves.

spread –

Transport barycenter νλ

Mass travels along the graph's edges.

spread –

Circle area is proportional to mass, on the same scale in both graphs. Spread is the root-mean-square distance of the mass from its own center, measured in grid steps.

An interactive demo highlighting the geometry of barycentric coding of measures on graphs.

Explanation →

Talk · slides in your browser A Barycentric Coding Model for Measures on Graphs FSU Topology & Geometry Seminar Papers & code Preprints and open-source packages Change point detection for time series, and barycenters of measures on graphs.

Latest posts