Under the supervision of Dr. Kevin Xu, I have worked on a variety of machine learning and data science projects from exploratory data analysis to deep learning.

Currently, I am working on creating a deep learning hybrid recommender system to improve matching between kidney donors and recipients intended to improve transplant outcomes.

During the 2019 spring semester, I worked on performing event detection on time-series traffic data for a challenge proposed at the 2018 NSF Algorithms Workshop. The features were a time series of inductive loop sensor reading from several locations. The goal of the challenge was to use this sensor data to identify if there was an event that occurred, such as an accident or an unusually high amount of traffic. For this challenge, I utilized several seq2seq models to classify the outliers. Models used include the Transformer Attention model, Deep Autoencoding Gaussian Mixture Model (DGAMM) and a custom VAE DAGMM model which improved upon the DGAMM model.

I also worked on developing an automatic identification system for maritime vessels for a challenge proposed at the 2018 NSF Algorithms for Threat Detection Workshop. The features given to us included vessel ID (label), speed of the ship, heading and the latitude and longitude of the last sighting. Sightings where veritable amounts of time, leaving time for the ship to travel potentially large distances between one sighting and the next. The goal of the challenge was to correctly identify which ship was which at each sighting of the ship. To tackle this challenge, I worked with Dr. Xu and a masters student in the lab to create a custom geo-temporal distance metric and model the ship travel using hierarchical clustering and Kalman filters.

During the 2018 fall semester, I worked on developing a model for skin conductance artifacting using convolutional neural networks and long-short term memory RNN’s using Keras and TensorFlow frameworks. I helped to develop dynetworkx by creating the SnapshotGraph data structure. dynetworkx is an open-source dynamic graph Python package based on the widely used networkx package. dynetworkx was released at the University of Michigan MIDAS Seminar series in September of 2018. In October of 2018, I traveled to Singapore to present our paper and poster at the HASCA Workshop at UbiComp 2018 for the research done during the summer of 2018.

Over the summer of 2018, I participated in the Sussex-Huawei Locomotion and Transportation Recognition Competition. The challenge provided smartphone sensor data with the goal of detecting classifying the mode of transportation the user was on. I worked on modeling this using a 1D CNN and an LSTM approach to attempt to capture the time series aspect of the data. As a result of this, our team was selected to present our paper’s finding at the HASCA workshop during UbiComp 2018. For funding over the summer, I participated in the Office of Undergraduate Research Undergraduate Summer Research and Creative Activities Program based on a research proposal I submitted.

I also manage the lab’s servers we use for training and testing algorithms via Puppet. I oversee managing users and software installations across the servers to ensure concurrency of applications. I have also assisted in developing an iOS app for collecting physiological data from participants in an upcoming trial.

During the summer of 2019 I worked at RoviSys in the Business and Industrial IT division. I installed and configured manufacturing networking and access control environments. Using VMware’s virtualization stack including ESXi, vSphere and vSAN, I configured and installed virtualized Windows domain environments including domain controllers, DNS, DHCP, WSUS relays and SMTP relays. I also created secure networking using paloalto firewalls, VLANs and network segmentation to create a three-zone DMZ networking environment with varying level of security. I also worked with Cisco switches to connect our physical devices with the rest of our network.

After completion of configuring the infrastructure, we then would create comprehensive documentation on the system including all detailing all physical devices in the system, software used, and custom configuration done to the infrastructure. After the documentation was complete, we would meet with the customer over several days and test that the system met their requirements fully.

I was chosen by the division management team to assist in creating a documentation standard for the division. This included creating templates for all common documents in the division and creating ways to efficiently and effectively fill them out as a lot of content was similar across documents.

My time at Rockford Construction was my first foray into a more enterprise environment. I primarily provided exceptional help desk service to over 300 employees, both in the office and in the field. During my down time, I learned about server grade virtualization and how to run an Active Directory environment. Additionally, since we were using MDT/SCCM for dynamic Windows build deployment, we needed a script to deploy printer drivers via MDT. I developed a PowerShell script automatically pull drivers from a share and install them during the MDT process. I also learned a lot about general sever management and how to develop a home-lab and deploy web services, along with proper general server hardening techniques.