Kara A. Ponder

I am interested in harnessing Artificial Intelligence to make the world a better place. I am currently a Senior Data Scientist at Noodle Analytics. Noodle works with other business using FlowOps to improve quality of manufactoring and improve supply chain flow in inventory.

Previously, I was a Research Associate in the Machine Learning Initiative at SLAC National Accelerator Laboratory. I focused on preparing for the Vera C Rubin Observatory's Legacy Survey of Space and Time (LSST) in the field of supernova cosmology. I used machine learning and artificial intelligence to prepare for the 100,000s of transient alerts per night. As a pipeline scientist for the LSST Dark Energy Science Collaboration (DESC), I helped to prepare the analysis pipelines for the 10+ year lifetime of LSST. I started and was leading the Supernova Machine Learning Topical Team in DESC to provide a space to discuss projects and new ideas for applications of ML to supernova cosmology.

Before SLAC, I was a postdoctoral researcher working at the Berkeley Center for Cosmological Physics as a Computational Data Science Fellow after graduating from the University of Pittsburgh with a PhD in Physics in 2017. I worked with Saul Perlmutter's group at UC Berkeley and LBNL on the Nearby Supernova Factory pipeline.

As a graduate student at the University of Pittsburgh, my research goals were understanding host galaxy correlations with supernovae for cosmology. I explored ways to improve parameter estimation with the likelihood function motivated by this correlation. I complimented other research of these correlations with NIR data, which I obtained in the SweetSpot survey that I was the lead graduate student for from 2014 to 2017.

Experience

  • Research Associate in the Machine Learning Initiative at SLAC
  • Berkeley Center for Cosmological Sciences Computational Data Scientist
  • PhD and Master of Science in Physics: University of Pittsburgh
  • Bachelor of Science in Physics and Astronomy: University of Georgia
  • Coding: Python, C++, pandas, sklearn, matplotlib, R, emcee, Django, PyRAF/IRAF, AstroPy, Django
  • Bayesian/Hierarchical Bayesian analyses
  • Dark Energy Parameter estimation
  • NIR Photometric Light curves of SNeIa for the SweetSpot Survey
  • Optical Spectroscopy of SN host galaxies

Highlights

Are Type Ia Supernovae in Restframe H Brighter in More Massive Galaxies?

The Astrophysical Journal, Volume 923, Issue 2, id.197, 34 pp. (2021) . This was my final first author paper. The plot shows the results of exploring correlations between host galaxy properties and supernovae observed in the NIR. We found evidence of a correlation between host galaxy mass and the brightness of the NIR-observed supernovae; however, the statistical significance of these correlations is at the 2-sigma level.

RESSPECT: Recommendation System for Spectroscopic Follow-up

This project uses active learning to make recommendations for which objects to spectroscopically classify given a limited number of telescope time verus the onslaught of possible targets that will come from LSST. The methods for the machine learning part of the active learning loop were published in Kennamer et al (2020) . I worked towards adding a cosmology-based metric to optimize cosmological constraints given possible follow up targets. The probabilty of classification, telescope availability, and this cosmology metric will be used to recommend spectroscopic follow up targets. Stay tuned for more papers from this project!

Photometric LSST Astronomical Time-Series Classification Challenge (PLAsTiCC)

PLAsTiCC was a machine learning challenge hosted at Kaggle with the aim to classify millions of lightcurves with a small, unrepresentative training set. I helped validate the simulations so that this competition could be leak free! I was also second author on the paper analysing the winning solutions.

Allegheny Observatory Public Lectures

"Exploring Dark Energy with the Large Synoptic Survey Telescope"
January 20, 2017

Incorporating Astrophysical Systematics into a Generalized Likelihood for Cosmology with Type Ia Supernovae

The Astrophysical Journal, Volume 825, Issue 1, article id. 35, 13 pp. (2016)

This plot, referred to by the authors as a "Butterfly" plot, illustrates a toy model looking at the distributions of supernovae. We built a framework to model systematics with a non-Gaussian likelihood that can remove bias with minimal precision losses.

Member of the Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST) Dark Energy Science Collaboration (DESC)

Member of Supernova working group
Co-lead of the Supernova Machine learning Topical Team
2015-2017, 2018-2020: member of Collaboration Council

Observations

WIYN 3.5 m on Kitt Peak using WIYN High-resolution InfraRed Camera (WHIRC) for SweetSpot.

WIYN 3.5 m on Kitt Peak using HexPak, a hexagonal shaped integral field unit mounted on the WIYN bench spectrograph for SweetSpot.

Bok 2.3 m on Kitt Peak for SDSS III Reverberation Mapping project.

Magellan Telescopes using Low Dispersion Survey Spectrograph 3 (LDSS-3): Slit spectrograph in Optical and Folded-port InfraRed Echellette (FIRE): Echelle mode NIR spectrograph. With LDSS-3, I helped observe the highest spectroscopically confirmed redshifted SLSNe.