I recently completed my Masters degree in Electrical Engineering and Computer Science at UC Berkeley with a focus on computer vision and started as a full-time software engineer at Facebook. Scroll down below to see my past industry and research experiences.
At Facebook, I worked on the Stories Creation Core team. My main project leveraged existing OpenGL infrastructure to extend swipeable filters used in Facebook Stories for iOS devices with post capture image enhancing filters. I also refactored the video trimming and tag tools' infrastructures to improve maintainability/modularity, and implemented redesign of doodle tool UI.
At Amazon Lab126, I worked in the Alexa Local Search Platform team. My project was to design and implement a service that would make calls to a NoSQL database to provide internal Alexa clients with requested data and perform query parsing to identify search keywords. I also performed unit tests and load tests to make sure the functionality and latency of the new system matched those of the existing system.
As an undergraduate researcher, I worked on the workshop paper High Accuracy Approximation of Secure Multiparty Neural Network Training which was accepted to AISys 2017. The goal of the paper was to explore linear approximations of common activation functions to improve accuracy of convolutional and recurrent neural networks in the context of efficient encryption. As a graduate student researcher, I worked on NBDT: Neural-Backed Decision Trees and the follow-up work SegNBDT: Visual Decision Rules for Segmentation . The goal of these papers was to construct decision-tree based classifiers using neural network weights to improve interpretability of neural networks while maintaining state-of-the-art accuracy for both image classification and semantic segmentation.
In the ADEPT Lab, I investigated entropy-based, non-uniform downsampling method to improve accuracy and reduce computation costs for the work Efficient Semantic Segmentation by Uncertainty-Based Downsampling
My masters thesis summarized the work done on Neural-Backed Decision Trees and includes a lengthy related works section on relevant background material. Click here to read my Masters thesis.