skills


Data Science

  • Expert in persistent homology, a method of feature extraction. 
    • Application: in Morphology of fly wings project Surabhi Beriwal, Ezra Miller, AT, and Dmitry Vagner will analyze data from Houle Lab to compare incidence of continuous and discrete morphological changes in wing veins across generations of fruit flies. 2-parameter persistent homology output to be analyzed is an invariant summary statistic for a persistence module; likely Hilbert function* with L_2 pseudometric. Responsible for data pipeline including advising multiple isolated software contributors (e.g. Beriwal, Vagner). 
  • Robotics work in GRASP Lab at Penn included using raw video, ultrasound, and binaural microphone data for vision and localization systems on humanoid robot. Advised by Dan Lee. 2010-2013. 
    • Designed and implemented a variation on a low-pass filter on ultrasound data for object detection. 
    • Used images segmented by learned color rules for object detection. 
    • Used binaural microphone data to compute direction from which another robot produced audible communication signal; sensor fusion of directional information from both microphone data and vision system via Kalman filter on a circle. Resulted in second author paper Active sound localization in a symmetric environment
    • Regularly trained vision system to learn colors in new lighting and environments. 
  • Mentored 1.5 year undergraduate project on approximating biological data (Surabhi Beriwal, Statistical analysis of fruit fly wing vein topology project). 2017-2018. 

Software

  • Significant past experience with: Matlab, C++,  C, Lua, Git, LaTeX, object-oriented programming
  • Some familiarity with: Macaulay2
  • Currently most comfortable with: Matlab, Git, LaTeX, object-oriented programming
  • Contributions to large, existing code bases:
    • Maintained and contributed to cross-platform humanoid robotics code base while part of the GRASP Lab at Penn. Most involved with computer vision, localization, and object detection systems. Code for robots written in Lua with some vision code written in C; debugging code written in Matlab. 2010-2013. 
    • Developed algorithms for HD video encoder during R&D internship at chip-design startup Immedia Semiconductor. Also independently wrote stand-alone data visualizer to help with algorithm development and debugging. C++. Summer 2010. 
  • Recent independent development:
    • Data visualizers for Morphology of fly wings project. Matlab, 2016-2017. Visualizers display:
      • Input fly wing veins plus the topological filtration used to computer persistent homology. Interactive via GUI: can adjust parameter values to show different filtered pieces; can show or hide how each parameter value contributes to respective filtered piece. 
      • Output in the form of preliminary approximations of Hilbert function* for persistence module. 
  • Mentored longterm undergraduate project in approximating Bezier curves with piecewise linear curves for Statistical analysis of fruit fly wing vein topology project. Focus on software best practices (modularity, design of data structures, appropriate documentation and commenting, unit testing, version control). Matlab. 2017-2018. 
  • Managing data pipeline with input experimental biology data from Houle Lab
    • Input data: a set of Bezier curves representing vein morphology for a fly wing. 
    • Near completion: Bezier curves approximated by piecewise linear curves. Includes computation of error using L_\infty metric. Surabhi Beriwal with heavy advisement by AT; advised by Ezra Miller.  
    • In progress: Computation of Hilbert function* for a 2-parameter persistence module constructed from piecewise linear fly wing. Dmitry Wagner with consultations from Pankaj Agarwal, Ezra Miller, AT. 
  • Undergraduate degree in electrical engineering (BSE, University of Pennsylvania 2013) with implementation focus on software for real-time embedded systems (including robots). 

Biology

  • Project Morphology of fly wings explores causes of discrete morphological differences between parents and offspring in structure of drosophila melanogaster wing veins.
  • 2 years of funding from an NSF Research Training Grant (RTG) in mathematical biology. 2015-2017. 
  • 2 years of weekly RTG math-bio colloquium and seminar series. Participants read seminar-speaker-provided background material and then presented and discussed that material at a colloquium. The seminar and lunch with the speaker followed later in the week. 2015-2017. 
  • Created and mentored 3 short-term undergraduate group projects in mathematical biology. Topics:
    • Statistics on the space of all evolutionary trees. In particular, how can we find an "average" for trees? 2017. 
    • How to construct evolutionary trees given gene sequence data. 2015. 
    • Persistent homology and applications: analyzing fly wings and brain arteries. 2017. 

Advising, Mentoring, and Interfacing with Non-Experts

  • Awarded L.P. and Barbara Smith Award for Excellence in Teaching in 2016 for 
    • "Consistent level of excellence" as primary instructor for Calculus 1 (Math 111L) at Duke over two semesters.
    • Significant organizing and outreach work for Southeastern Conference for Undergraduate Women in Math, 2015, 2016.
  • Mentored a 1.5 year undergraduate project as part of Morphology of fly wings project. Includes mathematical work and coding. Weekly 1+ hour meetings to jointly check in on and set short-and long-term goals, discuss difficulties and new material, and pair program. Student Surabhi Beriwal awarded Duke Mathematics Prize for Excellence in Research 2018 and met research requirement for Graduation with High Distinction in 2018, which requires noteworthy original results including significant new ideas beyond standard or established approaches. 
  • Created and mentored 7 groups of undergraduates and high school students (16 students total) in short-term group projects studying advanced topics in applied math. 2015-2017. 
  • Led year-long reading course with lower-level graduate student mentee. 
  • Significant experience presenting to non-expert audiences: 2012-2017
    • Presented biological information to mathematical audience for math-bio colloquium series (3 presentations, 20 minutes each). 2015-2017. 
    • Presented on advanced mathematical topics to undergrads (3 presentations, 20 minutes each). 2015-2016. 
    • Short course on advanced mathematical topics to high school students (5 hours of lectures). 2016. 
    • Over 20 outreach presentations for GRASP Lab. Usually led discussions and demos of humanoid robotics and sensing, but sometimes medical robotics (including demos of Da Vinci surgical robot) and techniques for human-robot communication like haptics. Commended by then GRASP Lab Director Kostas Daniilidis for contributions to the lab. 2010-2013. 

* The Hilbert function of a persistence module is a function from the parameter space to ℕ. It records the dimension of the vector space (the number of independent homology classes) at each parameter value.