Zack Carson

Academic Research, Gravitational Waves: University of Virginia

As a PhD candidate under the advisement of Professor Kent Yagi, the research I was involved with consisted of studying various aspects of astrophysics and general relativity via the detection of gravitational waves from the collisions of compact objects. This was split into two branches: constraining the supranuclear matter equation of state from the mergers of binary neutron star systems, and testing our current understanding of General Relativity via the collisions of black holes in binary systems.

Supranuclear matter equation of state:

Neutron stars in a binary system emit radiation in the form of gravitational waves as they lose energy and inspiral into each other. As the stars approach one another, they develop a tidal deformation in response to the companions strong tidal field, which affects the resulting trajectory. Information regarding this tidal deformation depends strongly on the underlying supranuclear equation of state, and becomes encoded within the gravitational waveform. By extracting such tidal information (from events similar to GW170817 detected by the LIGO collaboration), we place bounds on the nuclear matter parameters descriptive of the equation of state. We similarly use such tidal information to reduce the systematic uncertainties found in universal relations between various neutron star observables, allowing more accurate parameter extraction from future gravitational wave detections.

Testing general relativity

By instead studying the mergers of binary black hole systems (similar to GW150914 detected by the LIGO collaboration), we test the accuracy of Einsteins’ General Relativity. By utilizing Fisher Analysis techniques, we estimate the extraction efficiency of Parameterized Post-Einsteinian (PPE) parameters, which describe deviations from the General Relativity gravitational waveform entering at various orders of orbital velocity. By placing bounds on such parameters, we map them to the characteristic coupling parameters of various theories of modified gravity. By taking such an agnostic view, we can constrain the size of non-General Relativity effects relevant in the high-field, high-curvature regime of black hole mergers.

Orbital Analyst: United States Space Force

I currently work as an Orbital Analyst with the United States Space Force at the National Space Defense Center. Our mission is "Watch, Warn, Win", with the goal of watching allied and adversarial satellites to build a complete picture of Space Domain Awareness (SDA). By fusing orbital analysis with intelligence, we aim to maintain US space superiority while keeping all commercial, allied, and US assets safe from enemy disruption - allowing everyone safe and uninterrupted access to space.

Data Science: Dataminr

Previously, I worked as a Data scientist at Dataminr. Dataminr is the world's leading AI platform for real-time event and risk detection. Here, we detect the earliest signals of high-impact events and emerging risks from publically available information around, generating real-time alerts that address real-world problems. With our product, corporate and public sector clients (including big names such as The Washington Post, CNN, NYC Emergency Management, UPS, Shell, Citi, etc.) around the globe can obtain critical information first, respond with confidence, and manage crises more effectively. Recently Dataminr was ranked #5 on the Forbes AI 50 list, and we raised $475M on a $4.1B valuation.

• Analyzing internal application usage data,
• Testing various machine learning models on a variety of different sources,
• Creating and experimenting with annotation tasks on Amazon Mechanical Turk (AMT) for training machine learning models on various sources, and
• Investigating the detection (and alerting) of cyber-security vulnerabilities and threats towards clients using natural language processing (NLP) and computer vision (CV) machine learning.

Machine Learning, User Classification: TruU

TruU uses a unique combination of self-sovereign biometrics and behavioral biometrics to create the most frictionless access user experience for physical resources and computing systems. TruU works across the broadest range of enterprise systems requiring authentication including workstations, applications, VDI, VPN and physical security systems.

In short, by leveraging unique behavioral biometrics such as gait (walking), typing, and other behavioral patterns, TruU's product allows users across corporations to verify their identity by being themselves - allowing them to open secure doors and unlock computers without ID cards or passwords. Simultaneusly, any imposters will be easily detected by the models and will be locked out of everything entirely. This system not only drastically simplifies employees' lives, but it is also much more secure.

As a machine learning data scientist at TruU, I previously worked to implement machine learning models with the primary purpose to verify that users are who they say they are, based on various behaviorial biomeetrics they exhibit while going about their day. In particular, I worked on the following three models, all built in Python:

Gait Recognition:

The primary goal here was to identify users based on how they walk. As they move around with their phone, I begin by transforming the data into an invariant reference frame. This is the passed through a Contrastive Predictive Coding (CPC) machine learning model that has been trained on multitudes of gait data. This then allow me to compute risk scores on different users, based on if their current walk patterns matches how the model thinks they walk. Currently, this achieves an accuracy of around 98%!

Typing Recognition:

This model was used to identify users based on patterns recognized as they type on their keyboard. This is a metric-learning model which is trained on a large set of typing data, which can then determine whether given typing samples belong specific users or not, given an initial typing sample. Currently, this achieves an accuracy of 98%!

Hand Recognition:

Finally, this model was used to verify users based on a image of their hand. Just like above, this is a metric learning model which is trained on a large set of hand images, which can then determine whether given hand photos belong to specific users or not, given initial hand photos. Currently, this achieves an accuracy of 96%!

Game Development

In addition, I spend my spare time as an amateur game developer in Unity, making fun games as displayed in the Game Development tab above. In particular, my interests lie in making physics based games (especially astrophysics as per my research interested described below), and RPG's (Role Playing Games) of all kinds. See my game development tab above to experience the many games I've created!

Currently, I am working on a gravitational wave outreach video game, used to teach students about black hole and neutron star binary systems, and the resulting gravitational wave radiation. The player will then be tasked to navigate the intracacies of the resulting warped spacetime as gravitational waves pass through them. This project is sponsored by NSF in collaboration with Dr. Kent Yagi (University of Virginia).