About Me

 

Masters of Intelligent Information Systems student
 at Carnegie Mellon University, 
actively looking for a summer internship.  

Recently graduated from William and Mary,
completing a double major in Computer Science and Neuroscience
in three years,
while competing for the Division 1 gymnastics team.

Experience

A Pain Free Nociceptor: Predicting Football Injuries with Machine Learning

William & Mary Computer Science June 2020 – May 2021

I have been doing gymnastics since I was two, and knowing how detrimental injuries are in sports, I jumped on the opportunity to predict injuries in football. This started out as an REU project funded by the NSA and later turned into an Honors thesis advised by Gang Zhou. Amanda Watson from Gang Zhou’s LENS lab first introduced me to the project after her collaboration with the Athletic Department, which had collected daily sensor data on the football players alongside their injuries during the 2019 season. With more than half of the data missing due to various human factors, data interpolation methods were a big part of the project where we tested multiple different methods of interpolating missing data. Next, with the advice from Zhenming Liu, we analyzed the predictive power that parameters had in the training set and the test set, carefully selecting parameters that could predict injuries. Working on this project has given me a sense of how to process, visualize, and model data to extract information. While the predictive power for injuries within our data was fairly limited, working on this project furthered my interest in exploring the possible applications of data science.

Paper submission to Elsevier Smart Health: Accepted

Magneto: Joint Angle Analysis Using an Electromagnet-Based Sensing Method

William & Mary Computer Science April 2019 – June 2020

Last year, Amanda Watson had an idea of using a magnet and a magnetic field reader to track the relationship between the two for motion capture. As the magnetic field reader can have the same reading in multiple places around the magnet, we tested whether it is possible to fix the magnet and the magnetic field reader around the shoulder in such a way that only one joint position would correspond with a magnetic field reading. After modeling the system in python, we found that we could narrow down one reading to two possible joint locations, and proceed with real world implementation. After addressing the environmental magnetic field, we ran a user study to see how well a physical system would model the shoulder joint using neural networks. While it was able to predict shoulder angles correctly on individual users with up to a six degree error, it failed to scale across multiple users. When we collected data on the elbow joint, we kept the distance between the magnet and the field reader constant and found that the results could scale much better. However, for the shoulder joint, I wanted to keep the placement on the body constant, which brought in many variables due to physiological differences that I did not anticipate. This experience made it apparent how the data collection process and signal processing is a critical part of any data-driven project and how important careful data collection is, along with the understanding of the collection process. 

Patient pending
Citation: Watson, Amanda, et al. “Magneto: Joint Angle Analysis Using an Electromagnet-Based Sensing Method.” Proceedings of the 20th International Conference on Information Processing in Sensor Networks (co-located with CPS-IoT Week 2021). 2021.

Personal Projects

NLP & Crypto

Graduating college, I had noticed that Bitcoin was experiencing drastic day to day fluctuations. At the same time, I was looking for possible NLP related projects to play with on my own time. Hence, I stumbled across the question: Can twitter feed be used to predict changes in cryptocurrency prices? Seemingly so, as the price of cryptocurrencies is largely speculative, and most of the speculations that people make is based on what they see or read on the internet? Starting the project I ran across a paper that suggested that cryptocurrency trends can actually be predicted using tweet frequencies. I made a twitter scraper, and quickly modeled tweet frequencies to detect days when the price of bitcoin grows. As the model had a very low recall, I modeled other coins, yet many of the less popular coins did not provide conclusive results. As I delve more into this project, I hope to extract more complex information from tweets using NLP, and use that to improve the model.

Limited Vocab Story Generation

I first got interested in machine learning when making an app for learning languages. Growing up in a Russian community, I saw the barriers that language limitations place on people. I tried creating a natural language generator that would slowly build up the vocabulary used for generation. I initially hardcoded a generator, pairing adjectives that would work with nouns, pairing nouns with verbs, and embedding a grammatical structure, but this resulted in a very choppy and repetitive text. Next, I tried training an LSTM network but was not very satisfied with my result as the text failed to convey any meaning. Working on the project first gave me exposure to what machine learning is, the mathematical foundation for the field, and some exposure to the variety of neural network models being used. It was also the first time that I had seen a problem that had not yet been solved, sparking my interest to learn more about it that remains to this day.

Skills

Programming
 
  • Python
    • Pandas
    • Matplotlib
    • Numpy
    • Scipy
    • Keras
    • Tensorflow
    • OpenCV
  • Java
  • C
  • HTML/CSS
Advanced Math
 
  • Linear Algebra
  • Probability
  • Statistics

NLP

  • LSTMs
  • Sntwitter (tweet scraper)
  • Tokenization
  • Natural Language Generation

Other

  • Git
  • Linux
  • Signal processing
  • Data interpolation
  • Data modeling
  • ML model testing
  • Web development

Awards

  • NCAA All American on Parallel bars (2021)
  • CoSIDA Academic All American 3rd team (2021)
  • Graduated WM with Computer Science Honors (2021)
  • Jan 22 Tribe Leader of the Week (2021)
  • William & Mary School record on Highbar (2020)
  • 3 time provost award recipient (2019-2021)