About me

I am a postdoc at STScI (previously JHU) working with Josh Peek. I completed my PhD at Rutgers under the supervision of Andrew Baker. During my undergrad at Carnegie Mellon University, I worked with Rachel Mandelbaum. I use multi-wavelength observations combined with deep learning methods in order to study how galaxies grow and evolve.

Some links can be found in the upper right-hand corner. If you're interested in learning more about my machine learning work, be sure to check out my research blog.



Research interests and published works

Predicting optical spectra from galaxy imaging.

Predicting spectra from images

We introduce a novel neural network architecture with hybrid feature normalization in order to predict galaxies' SDSS spectra from Pan-STARRS imaging. We find that deconvolution operations in early layers of the network make optimization more robust.

Wu & Peek 2020, NeurIPS workshop
Interpreting why some galaxies are HI-rich and some are HI-poor.

Galaxy morphology and HI content

Using a deep learning approach, I estimated the neutral hydrogen (HI) gas-to-stellar mass ratio solely using optical image cutouts. Our predictions are generalizable, interpretable, and valuable for probing other properties that covary with HI and morphology.

Wu 2020, ApJ, 900, 142
Low-redshift analogs of Lyman break galaxies.

Lyman break galaxy analogs

We studied nearby analogs of distant Lyman break galaxies (LBGs) using resolved VLT/SINFONI near-infrared spectroscopy. LBG analogs appear to have hard ionizing spectra, which may evidence massive star binaries, as well as reservoirs of warm molecular gas heated in photo-dissociation regions.

Wu et al. 2019, ApJ, 887, 251
An image of the massive mergiving cluster, El Gordo.

Probing metallicity with CNNs

We trained a convolutional neural network to predict the gas-phase oxygen abundance, or metallicity, of a galaxy using only optical gri imaging. It performs so well that we can reconstruct the mass-metallicity relation, using CNN predictions, with zero addition scatter!

Wu & Boada 2019, MNRAS, 484, 4683
SDSS galaxies with metallicity predictions.

Galaxies in massive clusters

We investigated the dust, gas, and star formation properties of galaxies residing in the most massive clusters in the z ~ 1 Universe with Herschel and ALMA observations. We found unexpectedly massive cold gas and dust reservoirs, but low star formation rates.

Wu et al. 2018, ApJ, 853, 195
A picture of a MeerKAT antenna.

The LADUMA Survey

Looking At the Distant Universe with the MeerKAT Array, or LADUMA, is an ongoing survey that will be the first to detect galaxies in neutral atomic hydrogen emission out to z = 1.4. I've worked on calibration, source-finding, and obtaining optical spectroscopy for LADUMA.

Blyth et al. 2016