code
Some of the open-source projects I work on. See my GitHub page for a more complete list.
A software flowchart for the HERA analysis pipeline.
The HERA Analysis Pipeline
My collaborators and I have developed the main HERA calibration and analysis pipeline. I’ve worked specifically on custom software for data quality metrics, calibration and systematic treatment, and power spectrum estimation, with a particular emphasis on systematic models. All software is unit tested, peer-reviewed, and has documentation in each respective repository for getting started with the code.
Cosmological image segmentation with convolutional neural networks.
CNNs for Cosmological Simulations
A PyTorch implementation of convolutional autoencoders with application to image segmentation of cosmological radiative transfer simulations. Once trained, this model can be used to quickly determine ionized regions of the universe given an evolved matter density field. This is important because radiative transfer is generally the most expensive part of large scale cosmological simulations. The ability to accelerate this step can dramatically improve the size of models we are able to run with a given amount of time and compute.
Posterior inference of a Cosmic Dawn simulation, accelerated with emulators.
Cosmic Dawn Emulators in Python
Surrogate models, or emulators, are ways to approximate simulations that are otherwise too time intensive or computationally expensive to run directly. They allow us to produce robust, Bayesian parameter constraints on sophisticated models of cosmology and astrophysics. In Kern et al. 2017, we created a Python-based emulator framework for relating 21cm intensity mapping survey data to sophisticated models of Cosmic Dawn and the Epoch of Reionization.
The generalized Python emulator code can be found in the emupy software. The code supports polynomial, Gaussian Process, and neural network based surrogates, and contains minimal working examples.