Image Recognition Using Convolutional Neural Network in TensorFlow
This project involved using transfer learning to train a CNN for the purpose of classifying images of birds into respective species. The images came from a publicly available dataset on Kaggle, and were used to train a CNN build on the popular Xception model using the TensorFlow framework for Python. The model is capable of classyfing birds into 175 different species with 98% accuracy. The source code and testing code are available in the GitHub repo, and a TensorFlow Serving Docker image pre-loaded with the model is available on DockerHub. With the Docker image, the model can be deployed most anywhere, can be queried through a REST API, and supports versioning.
Masters Thesis - KU Leuven June 2019
Reliability Analysis of Mechanical Equipment in a Cement Production Plant
In June 2019, I completed work on my Master’s thesis as the culmination of my two-year program at KU Leuven. As indicated by the sheer length of my document, the nature of my project enabled me to approach a solution from an enormity of angles. I was fortunate enough to be tasked with a concrete(pun absolutely intended) problem with a trove of historical data at my disposal. Using knowledge from my previous Generalized Linear Models, Neural Networks, Machine Learning, and Survival Analysis courses, I implemented a widely applicable methodology that I’m proud of.
In May 2018 as part of the statistics curriculum I carried out considerable research and performed an independent meta analysis on a topic of my chosing. Being an endurance athlete, I had always wondered about the science and facts behind alleged recovery-aids. This project seemed the perfect opportunity to finally separate fact from fiction.
Privacy and Big Data
The culmination of the Privacy and Big Data course at KU Leuven involved taking what we had learned and performing a Privacy Impact Assessment of our own. I chose to examine Strava, a hybrid platform at the intersection of exercise/training log and social network. Despite being a long time user(made easy by cross-platform activity syncing) I learned a great deal about the amount of value generated from aggregated activity data. Regardless of what I discovered, I still continue to use the platform in the same semi-passive manner.