I'm a postdoctoral researcher working in astrophysics.

Currently I use machine learning techniques to better understand how gravitational lensing can tell us about dark matter.

In general, I'm interested in any application of statistics and computation to interesting problems, especially if it involves machine learning, Bayesian inference, or high-performance computing.

CURRENTLY

Dark Matter Group

Max Planck Institute for Astrophysics

Garching, Germany

More here

Astrophysics Group

Imperial College London

UK

PREVIOUSLY

Talks

UPCOMING

Debating the potential of machine learning in astronomical surveys 2

IAP Paris, France / Flatiron Institute, New York, USA | Nov/Dev 2023 | link

PAST

DSU 2023: 17th international workshop on the dark side of the universe

EAIFR, Kigali, Rwanda | July 2023 | link | slides

IAU Symposium 381: strong gravitational lensing in the era of big data

Otranto, Italy | June 2023 | link

Nature of dark matter on small scales

Yale seminar series, over zoom | Sep 2023

SHARP collaboration meeting

Piemonte, Italy | June 2023

Origins data science workshop

Origins Excellence Cluster, Garching, Germany | Jan 2023 | link

Munich dark matter meeting

MPA, Garching, Germany | Dec 2022 | link

AI goes MAD workshop 2022

IFT, Universidad Autónoma de Madrid, Spain | Jun 2022 | link | slides

Euclid strong lens science working group meeting

Marseille, over Zoom | Dec 2022

Debating the potential of machine learning in astronomical surveys

IAP Paris, France | Oct 2021 | link | video below

Projects

Using our machine learning method developed previously, we examined the effect of angular complexity in the lens model on dark matter substructure detection. Using CNNs trained on data with and without angular multipole perturbations, we found that small angular perturbations in the lens can cause a significant rate of false positive substructure detections.

ANGULAR COMPLEXITY IN STRONG LENS SUBSTRUCTURE DETECTION

FAST SENSITIVITY MAPPING AND EUCLID FORECASTS

When strong lensing is used to detect dark matter subhaloes, it is essential to know the sensitivity of the data, that is, what's the smallest DM subhalo that one could detect, if it were there. This information turns a set of subhalo detections and non-detections into an inference on the true dark matter model of the universe.

We developed a machine learning method to map the sensitivity of strong lens observations. This was previously a very expensive and time consuming part of the analysis. With fast and cheap sensitivity mapping, we can learn much more about the way that dark matter subhaloes effect strong gravitational lenses, allowing us to better focus our detection efforts.

Using our new method, we computed sensitivity maps for 16k simulated Euclid strong lens images, allowing us to forecast the number of detections we expect in that survey.

GALAXY MASS PROFILES FROM STRONG GRAVITATIONAL LENSING

For my Ph.D. work, I developed the theory for constraining galaxy mass profiles with strong lensing data, when both image position and image flux information are combined. If the lens is elliptical, the data are sensitive to the structure of the galaxy interior to the lensed images. The structure in this region is sensitive to the galaxy formation history and the interplay between dark and baryonic matter.

The model developed in this work has been employed elsewhere, including in the detection of the largest known supermassive black hole.

See a full list of my publications on ADS