I'm a Master's student at Stanford University studying Electrical Engineering and Computer Science under Prof. Stephen Boyd. In my professional life I am deeply interested in the design of emerging technologies, especially in the area of artificial intelligence, to develop lasting communities in economically and socially sustainable ways. In my spare time, I am an amateur photographer and an avid video gamer; I've also been known to spend whole afternoons outside reading a good sci-fi novel. Always happy to discuss any shared interests!
Advisor: Stephen Boyd
Relevant Courses: Machine Learning, Probabilistic Graphical Models, Parallel Computing, Mining Massive Datasets, Principles of Computer Systems
Advisor: Stephen Boyd
Relevant Courses: Convex Optimization, Information Theory, Stochastic Signal Processing, Linear Dynamical Systems
I spent the summer of 2019 exploring the field of quantitative finance in London. In the first part of my internship, I augmented the company's existing data processing infrastructure, which was designed for continuous signals, to capture and analyze data about discrete financial events in realtime. After completing that project, the remainder of my time was spent developing an neural autoencoder to discover latent risk factors. The final model that I presented exhibited a 10% increase in profits when compared to the existing model on historic data that it had not been trained on. Finally, I spent the last few weeks of the internship creating black box analytics tools for machine learning models to facilitate interpretability of the model I developed and others used in the company.
Working with a Ph.D student in the SNAP group under Prof. Jure Leskovec, I developed a graph convolutional neural network architecture for the problem of subgraph isomorphism matching. This is a foundational NP-complete problem in theoretical computer science, and is of interest in many domains including biology (protein interaction networks), search (knowledge graphs), and chemistry (structure matching). We developed a model which identifies queries that it is trained on with near perfect accuracy, and generalizes well to unseen queries. We are currently working on our paper.
At Apple, I was part of a small research team looking to develop secure decentralized authentication protocols using blockchain technology. My main project was to build the complete backend for a proof-of-concept where a client using a phone could authenticate to a remote server backed by a proof-of-stake blockchain running on AWS. On this project, I also served as a project lead working to facilitate the work of another team member working on the front end interface and coordinate his work with mine as well as our manager's vision. After presenting our demo to the VP overseeing our division, I was invited to continue working part-time during the school year. In this time, I created a SQL generation tool which could convert natural language descriptions of entities and relationships into a formal database schema, and then create an app to interact with said database.
Under Prof. Andrew Ng, I worked in a team with three graduate students to develop a 3D image segementation network for the diagnosis of cerebral aneurysms. We developed a tool for radiologists which would highlight, in realtime, possible aneurysms on a CTA scan using inferences from our model. When using this system on new CTA scans, practicing radiologists from the Stanford Hospital demonstrated statistically significant improvements in accuracy, recall, and interrater agreement. You can find our paper, published in the Journal of the American Medical Association, here.
Invited by a professor from one of my classes, I spent my first summer at Stanford doing research on efficient implementations convolutional neural networks for the US Army. Using iterative sparsification and pruning methods, I was able to reduce the size of a object detection network (intended to be used as part of an autonomous vehicle project) 100x while maintaining performance, which sped up both training and inference times by 50x. For my work, I was awarded the program's best project award.
I worked for a summer in high school as a member of a bioengineering research lab at the University of Cincinnati, in a group developing a new delivery mechanism for stroke treatment drugs (rt-PA) using echogenic liposomes and ultrasound. My work focused on fine-tuning a standard operating procedure for a machine which measured the size of particles in solution.
Click on a project to access the corresponding Github repository.
Recurrent neural network in TensorFlow for learning codes for communication over arbitrary noisy channels. Our model achieves performance within a small factor of the Shannon limit for several common noise models even with unknown noise, and we extend the learning scheme to discrete channels by using dithering to create a continuous approximation.
Siamese neural network architecture using PyTorch which matches textual queries to relevant segments of videos from a preprocessed database. The architecture uses only semantic content from the video to match against the query, rather than relying on human annotations or tags.
A python framework for building simple neural networks with automatic differentiation. Supports many common activations functions (relu, sigmoid, tanh), and optimizers.
Army High Performance Computing Research Center
National Merit Scholarship Corporation
US Presidential Scholars Program
Copyright © All rights reserved | This template is made with by Colorlib