Laurence Perreault-Levasseur

Laurence Perreault-Levasseur is a cosmologist who develops machine learning methods for analyzing strong gravitational lenses and constraining cosmological parameters. After a PhD at the University of Cambridge in 2015 on inflation, she held postdoctoral fellowships at Stanford's KIPAC and, from 2018, at the Flatiron Institute. In 2017, she coauthored a Nature study showing that convolutional neural networks analyze gravitational lenses about ten million times faster than traditional methods. She joined the Department of Physics at the Université de Montréal and Mila in August 2019, then coauthored work on score-based priors in 2022 and the Caustics simulation package in 2024. She is an associate professor at the Université de Montréal, an associate member of Mila and the Canada Research Chair in Computational Cosmology and Artificial Intelligence.

Now at: Université de Montréal, Quebec Artificial Intelligence Institute, Simons Foundation and Perimeter Institute

Timeline

By 2015

A Cambridge PhD on inflation

After bachelor's and master's degrees at McGill University, she completes a PhD in applied mathematics and theoretical physics at the University of Cambridge in 2015. Her doctoral work applies open effective field theory methods to the formalism of inflation.
2017

Neural networks for gravitational lenses

As a postdoctoral fellow at Stanford's Kavli Institute for Particle Astrophysics and Cosmology (KIPAC), she coauthors a Nature study in 2017 with Yashar Hezaveh and Philip Marshall. Their convolutional neural networks analyze gravitational lenses about ten million times faster than traditional methods, handling 100 systems in about a second on a single graphics processing unit.
2018

Postdoc at the Flatiron Institute

A second postdoctoral fellowship takes her to the Flatiron Institute in 2018, where she is a Flatiron Fellow at the Center for Computational Astrophysics.
January 2019

Recurrent inference machines for lensing

In January 2019, she is one of nine authors, with Warren R. Morningstar, of a paper that uses a Recurrent Inference Machine to reconstruct undistorted images of background sources in strongly lensed systems. A convolutional neural network that predicts the lens mass distribution directly from telescope images completes a fully automated reconstruction of the source and the foreground mass.
August 2019

Joining Université de Montréal

In August 2019, she joins the Department of Physics at the Université de Montréal, with a joint position at the Quebec Artificial Intelligence Institute (Mila). Her research there centers on cosmology and on new machine learning techniques for data analysis and for constraining cosmological parameters.
November 2022

Score-based priors for lensed galaxies

A paper from November 2022, coauthored with Alexandre Adam and others, uses a score-based model as the prior for inferring undistorted images of background galaxies in strong gravitational lenses. Adding the likelihood score to the prior score and solving a reverse-time stochastic differential equation produces independent posterior samples that model the data almost down to the noise level.
November 2024

The Caustics lensing package

The Journal of Open Source Software publishes Caustics in November 2024, a Python package for accelerated strong gravitational lensing simulations. She is a coauthor, with Connor Stone, Alexandre Adam, Yashar Hezaveh and others.
As of October 2026

Associate professor in Montréal

She is an associate professor at the Université de Montréal and an associate member of Mila, and a visiting scholar at the Flatiron Institute in New York City and the Perimeter Institute in Waterloo. She holds the Canada Research Chair in Computational Cosmology and Artificial Intelligence. In December 2024, she works with a team of around twenty researchers, mostly PhD and postdoctoral students, on deep learning methods for strong gravitational lenses ahead of the Legacy Survey of Space and Time at the Vera C. Rubin Observatory.