Nicholas Kern
Research Fellow | astrophysics & cosmology, radio astronomy, machine learning
Ann Arbor, MI, USA
I am a NASA Hubble Fellow based at the University of Michigan. I am a data-oriented researcher working at the interface of high-redshift astrophysics & cosmology, radio astronomical observation and machine learning. My research goals are to use next-generation radio telescopes to map the high-redshift universe with unprecedented statistical precision, enabling us to tap into a trove of currently unharnessed cosmological information. To do this, I develop novel algorithms and inference frameworks to robustly characterize weak signals in noisy, large-scale datasets. Some of the broad questions my research aims to address are:
- How did the first stars, black holes, and galaxies form, and how did their radiative feedback impact the surrounding primordial hydrogen?
- How do high-redshift observations of the universe fit with low-redshift observations, and how can they stress-test our cosmological model?
- How can we robustly and optimally extract weak cosmological signals from noisy and systematics-contaminated data, and how can advances in machine learning / AI accelerate and enhance this process?
I leverage data from powerful radio telescopes, and design novel, ML-driven analysis frameworks for addressing these questions. Some of my current work is in developing the first fully end-to-end and differentiable Bayesian forward model for 21cm line intensity mapping telescopes that will reveal the universe’s structure at early times. I am also working on understanding how generative models can be reliably applied to statistical inference problems within the physical sciences, with a focus on generalization and model misspecification.
Previously, I was a Pappalardo Fellow at MIT. I received a PhD in astrophysics from UC Berkeley in 2020, and a BS in physics and astrophysics at the University of Michigan in 2015.
nkern@umich.edu
github.com/nkern
google scholar
linkedin/nicholas-kern