Research interests

Most of my research is about optimal transport: the mathematics of moving one distribution of mass onto another as cheaply as possible. It turns out to be a very natural way to measure how far apart two probability distributions are, which is why it shows up across machine learning, statistics and economics.

I work on its algorithmic side, and in particular on making it usable on real data. Classical optimal transport struggles when the data has many dimensions, and it is easily thrown off by noise and outliers. A lot of my work adds structure to the problem, through low-dimensional projections, convexity and regularity, so that optimal transport stays reliable and fast to compute. I have used these ideas in machine learning and in economics.

More recently I have been working on time series and forecasting and on operations research, where the same questions of structure, regularity and efficient computation keep coming back.

See my work come to life in interactive demos →

Publications

Talks & tutorials