About
Projects
Iowa Center for Research by Undergraduates Fellow 2020
To study the decision boundaries of deep learning networks, adversarial attacks are constructed to fool M models while defending n models for varied M and n in this project. Being able to construct these attacks shows that deep learning networks behave differently despite identical training and architecture, providing insight into how deep learning classifiers behave and why they are susceptible to these attacks. The project features a secondary focus in detection of adversarial attacks in an MRI image reconstruction network. If biological data corruption follows a specific pattern where detection is possible, prevention of failure to adversarial inputs is then implied. To view the code, wherein additions are still being made, visit the Github repository below.
GitHubIowa Center for Research by Undergraduates Fellow 2019
To compare communication between populations that were supporters/dissenters of a particular candidate, a network of words was constructed around a particular candidate's name using word embedding (Word2Vec). For visualization purposes, only the top ten closest words were considered to the center word, and then the top three words closest to those words were displayed in a connected graph. The data used to create these networks of communication flow were 2 Terabytes of Twitter data collected over the course of the 2016 election cycle. To view the code or an example network, visit the Github repository below.
GitHubResearch Assistant at the Blumberg Lab of Brain Sciences
Primarily, I used neural networks for classification of twitch behavior in brain wave data using MATLAB. Secondly, I constructed a automated data collection tool for the lab. This project features the use of four magnetometers placed below each of the limbs of the mice used in experiments. As small magnets are attached to the mice's limbs (which are dangling off a platform suspended above the magentometers), the location data can be collected without the need for manual video tracking. The code, as well as a system diagram, can be viewed at the Github repository below.
GitHub