After seeing an incredible online resume, I was inspired to create my own. I really like the concept of using a website as a resume because it allows me to showcase my programming abilities as well as my creative side, and I believe it paints a more complete picture of who I am than any paper resume could. I elected to build this website in HTML, CSS, and JavaScript with JQuery instead of using a website builder because I wanted an opportunity to learn a bit of web development and gain an understanding of how the online content we consume every day is styled and delivered to us.

Machine learning techniques for 3D facial modeling can be challenging due to a lack of high-quality 3D data. Many image datasets used for deep learning have thousands or even hundreds of thousands of images. However, high quality 3D face datasets generally contain a few hundred image/model pairs. The goal of this project in my Physics Based Computer Vision course was to use a classic 2D face dataset (CelebA) along with the power of differentiable rendering (with help from an open-source Facebook AI Research renderer) to train a network that could accurately infer depth information about a face without having access to any ground-truth depth data. I worked on a team as the primary neural network architecture designer and code developer.

During the entirety of my Semiconductor Characterization course, I was searching for ways to apply machine learning concepts to the course material. For the final project, I found a team interested in pursuing my idea of a diode characterization network. I developed a network to infer material parameters about the diode, including doping concentration, substrate material, and contact material. With a few exceptions, the classifications and regressions were very accurate, and in fact the contact material was classified correctly more than 99% of the time. I wrote a python script to generate 40,000 Silvaco ATLAS simulation files with various material parameters and run those simulations in a distributed manner across 16 servers.

In my Python for Rapid Engineering Solutions course, we briefly discussed the concept of simulated annealing in the context of the Traveling Salesman Problem. I was curious to see if simulated annealing could be applied to other types of optimization problems. I discussed it with my Advanced Analog Circuits professor, and he told me he hadn't heard of anyone using this technique for circuit synthesis, but that he was more than happy to guide me in an independent study on the topic the following semester. This project showed me the strength of simulated annealing as a method for circuit synthesis, and taught me a lot about efficient circuit simulation and data processing.

This was my first machine learning project. I was in a group with two other students working to improve upon a machine-learning based image upsampling algorithm. The team spent most of our time discussing various techniques that could be used, and experimented with multiple network architectures. This project exposed me to many scientific papers, machine learning ideas, and deep learning techniques, and was the catalyst that made me quickly fall in love with machine learning.