- Airborne Collision Avoidance System X (ACAS X)
- MIT Lincoln Laboratory - Group 42
- Software for generating airborne collision avoidance advisories for manned and unmanned aircraft
- Logic formulated as a Markov decision process
- Computational strategies employed for handling large state space, including
- Multilinear interpolation
- Parallel computing
- Contributed visualization tool for analysis of policy evolution through the course of the value iteration algorithm
- C++, Julia, MATLAB
- ClutteredEnvPathOpt.jl (repo)
- Rice University - with Illya V. Hicks, Joey Huchette and Miles Olson
- Package for solving optimal path planning problems in cluttered environments for robots and drones
- Implemented the independent branching scheme to formulate obstacle avoidance disjunctive constraints, including an algorithm for obtaining the necessary biclique covers on a special class of graphs
- Includes infrastructure for creating obstacles, generating obstacle-free regions, and constructing the associated graphs
- Julia
- Audubon_F21 (repo)
- Rice University - Data to Science (D2K) Lab
- Package for identifying and censusing various colonial waterbird species from UAV imagery of nesting islands
- Sponsored by Houston Audubon for real-world deployment in their waterbird monitoring studies
- Employs a Faster R-CNN object detector with a ResNet-50 feature pyramid network backbone, implemented with Detectron2
- Capable of detecting 10+ species, with the 3 most populous (constituting over 70% of the population) being detected with an AP of over 90% (at an IoU of 0.5)
- Led experimentation of a customized implementation of Faster R-CNN utilizing a DenseNet backbone
- Bayesian hyperparameter optimization used for selection of learning rate and decay factor
- Data augmentation techniques for minority species include horizontal and vertical mirrorings, 90 degree rotations, and random brightness and contrast adjustments
- Python
- Forecasting Yearly Battery Replacements
- Rice University - Data to Science (D2K) Lab
- Mentored a team of students on a capstone data science project focusing on forecasting yearly battery replacements for LivaNova’s vagus nerve stimulator medical device
- Team employed survival analysis to generate battery durations for implants for both existing patients and new patients
- Existing patients: Survival function modeled with a Log-normal Accelerated Failure Time model
- New patients: Survival function modeled with a Kaplan-Meier estimator
- 1st Place in the Fall 2022 D2K Showcase
