About
I started my career in aerospace engineering. While a PhD candidate at Stanford, I studied artificial intelligence and developed neural network architectures for scientific simulations. During the pandemic, I ran a nonprofit to help fight covid with decentralized technology. I care about aligning technology with human rights and values.
Highlights
Slowing the Spread of Infectious Diseases Using Crowdsourced Data
Pandemic mitigation is fundamentally an optimization problem. Mobile technologies can provide instantaneous and high accuracy exposure alerts, even between strangers at low social and economic cost. We propose novel technology that performs automatic, anonymous, and decentralized exposure alerts using Bluetooth proximity networks. CEN/TCN Protocol White Paper from March 20, 2020. paper
Fast Neural Network Predictions from Constrained Aerodynamics Datasets
The cluster network automatically breaks problems into parts while learning to make aerodynamics predictions from very few examples. The datasets are designed to be typical of aerospace industry datasets, which are highly constrained by the number of examples. "Fast Neural Network Predictions from Constrained Aerodynamics Datasets." AIAA Scitech Conference Paper. arxiv link paper poster github
Make your own aerodynamic artwork easily in your browser (works on mobile)
There are GPU-accelerated fluid dynamics apps to play with now. I wanted one that allows you to upload your own images as dye ink to create aerodynamic artwork, so I modified a repository to easily make flowing art. Play here. It's free to play and make aero/fluid dynamic art. Works on desktop and mobile. github Example artwork on github.
A demo webpage rating button to reward content creators and build an ad-free ecosystem
This is a browser extension that lets you rate any webpage and an associated search engine with confidence scores based on positive and negative valence. I used the same scoring algorithm as reddit's top score plus noise to help elevate new webpages. This is an alternative to click-based models to help combat the excessive rewarding of polarizing content. A paid version could reward content creators to help reduce reliance on ads. Demo at linkUpvote.vote.
Reducing Transonic Flutter in Turbomachinery with Local Positive Camber
For flutter caused by a transonic shockwave, adding local positive camber to the blade design near the predicted shockwave location is found to eliminate flutter with negligible impact on aerodynamic efficiency. "Flutter-resistant transonic turbomachinery blades and methods for reducing transonic turbomachinery blade flutter." linkHoneywell Aerospace Patent.
Nonprofit Work
In February 2020, I started and began to run Covid Watch, a nonprofit with a mission to improve the privacy of the contact tracing apps that were being released in response to the pandemic.
We were the first team in the world to publish a white paper, develop, and open source a decentralized exposure notification protocol using Bluetooth communication in March 2020. If an app implements the security protocol, it ensures that data is stored locally on personal devices, and stays anonymous throughout the notification process. The system is designed to prevent corporations and governments from using the apps for the centralized collection of personally identifiable information. Our TCN Protocol received significant news coverage and was followed by the development of similar privacy-preserving protocols in early April like DP-3T, PACT, and Google/Apple GAEN.
In April 2020, our volunteers and nonprofit staff helped develop an early data rights framework for exposure notification, and released a free and open source mobile app with development costs funded by prizes and donations. In August 2020, we collaborated with the University of Arizona on research to improve the estimation of infection risk from decentralized Bluetooth data to better inform private quarantine recommendations.
At the end of 2020, the Covid Watch nonprofit closed, but the open source Covid Watch app continues to be implemented for universities and public health departments by WeHealth, a public benefit corporation.
Download WhitepaperIn the News
VentureBeat
Surveillance, AI, and saving lives top agenda at coronavirus conference. Khari Johnson. April 2nd, 2020.
NBC News
Behind the global efforts to make a privacy-first coronavirus tracking app. David Ingram and Jacob Ward. April 7th, 2020.
WIRED
Clever Cryptography Could Protect Privacy in Covid-19 Contact-Tracing Apps. Andy Greenberg. April 8th, 2020.
Stanford News
Stanford researchers help develop privacy-focused coronavirus alert app. Tom Abate. April 9th, 2020.
American Civil Liberties Union
ACLU White Paper – Principles For Technology-Assisted Contact-Tracing. Daniel Kahn Gillmor. April 16th, 2020.
Linux Foundation Public Health
Digital Contact Tracing and Alerting vs Exposure Notification. Harper Reed. April 22nd, 2020.
San Francisco Chronicle
Latest weapon in tracing and tracking coronavirus infections: your smartphone. Carolyn Said. April 22nd, 2020.
Associated Press
Here come COVID-19 tracing apps - and privacy trade-offs. Matt O'Brien and Christina Larson. May 5th, 2020.
Marginal Revolution
Emergent Ventures prize winners, third cohort. Tyler Cowen. May 18th, 2020.
Johns Hopkins University Press
Digital Contact Tracing for Pandemic Response: Ethics and Governance Guidance. JHPEGDCT and Jeffery Kahn. May 26th, 2020.
TechCrunch
Demonstrating 15 contact tracing and other tools built to mitigate the impact of COVID-19. Andy Moss, Connor Spelliscy, and John Borthwick. June 5th, 2020.
The Wall Street Journal
America Is Reopening. Coronavirus Tracing Apps Aren't Ready. Rolfe Winkler. June 22nd, 2020.
Arizona Public Media
NPR | Arizona Illustrated | Reopening Safer. Bryan Nelson. August 3rd, 2020.
ABC News
Smartphone app alerts University of Arizona students if they are exposed to COVID-19. Teri Whitcraft, Matt Gutman, and Alyssa Pone. August 25th, 2020.
USA Today
COVID-19 tracking apps, supported by Apple and Google, begin showing up in app stores. Rob Pegoraro. August 26th, 2020.
Project Syndicate
How (Not) to Fight COVID-19. Peter Singer and Joanna Masel. September 3rd, 2020.
Voice of America
New COVID App Gives Health Officials Dials to Send Tailored Alerts. Michelle Quinn. October 5th, 2020.
University of Arizona News
How a Smartphone App and Contact Tracing Helped Keep UArizona Open and Curb COVID-19 Spread. Mikayla Mace. December 16th, 2020.
Reason Podcasts
This App Protects Privacy While Tracing Covid-19 Infections. Nick Gillespie. May 6th, 2020.
KJZZ
University Of Arizona Alum Develops Confidential Contract Tracing App. Lauren Gilger. July 22nd, 2020.
More news...
Media
Tracking COVID-19 Using Crowdsourced Data. Stanford HAI's COVID-19 and AI Virtual Conference. April 1st, 2020.
The Role of Technology in Stopping the Spread of Covid-19. Tony Blair Institute for Global Change. April 30th, 2020.
Warning Tools for COVID Exposure: Telephone Town Hall. Sen. Glazer Town Hall. May 21st, 2020.
Decentralized exposure alert protocols for protecting communities from COVID-19. Toronto Machine Learning Series (TMLS). May 28th, 2020.
Approximating Solutions to Fluid Dynamics Problems & A Mobile App Intervention for COVID-19. Department of Energy CSGF Program Review. July 14th, 2020. link
The University of Arizona launches contact tracing app 12 News. August 20th, 2020.
Research
I originally worked in aerospace engineering, completing my M.S. at the University of Arizona. While at Honeywell Aerospace, I invented and patented a method for reducing the risk of transonic flutter in turbomachinery, improving the safety of next generation jet engines. After spending several years in industry, I decided to pursue a PhD at Stanford supported by a Department of Energy Computational Science Graduate Fellowship (DOE CSGF).
At Stanford, I focused on courses in computer science and machine learning. I developed two novel methods for faster and more accurate reduced order models and I designed neural network architectures for approximating the solutions to partial differential equations by breaking them into parts.
Space-local reduced-order bases for accelerating reduced-order models through sparsity
A space-local reduced-order basis is developed, which introduces sparsity in projection-based model order reduction (PMOR) by partitioning the computational domain rather than, or in addition to, the solution manifold. Acceleration factors of 1.5 relative to traditional PMOR and CPU time speedup factors of several orders of magnitude relative to high-dimensional models are shown for two computational fluid dynamics problems. IJNME Conference Paper, November 2022. paper
Quantifying SARS-CoV-2 infection risk within the Google/Apple exposure notification framework to inform quarantine recommendations
Most Bluetooth-based exposure notification apps use three binary classifications to recommend quarantine following SARS-CoV-2 exposure. Instead, we model uncertainty in the shape and orientation of an exhaled virus-containing plume and in inhalation parameters, and measure uncertainty in distance as a function of Bluetooth attenuation. U. Arizona and Covid Watch Risk Analysis Paper, June 2021. paper
Fast Neural Network Predictions from Constrained Aerodynamics Datasets
The cluster network automatically breaks problems into parts while learning to make aerodynamics predictions from very few examples. The datasets are designed to be typical of aerospace industry datasets, which are highly constrained by the number of examples. "Fast Neural Network Predictions from Constrained Aerodynamics Datasets." AIAA Scitech Conference Paper. arxiv link paper poster github
Two Architectures with LSTMs Inspired by Cortical Columns
Two architectures inspired by cortical columns are compared. The baseline cluster network performs better than the LSTM cluster network because time doesn't need to be treated differently from space when the datasets are fully resolved in both. "A neural network for breaking time dependent problems into parts." Stanford CS379C: Computational Models of the Neocortex Class Project. paper github
An Autoencoder / Image Captioning Model and an Architecture Search for CFD Predictions
An image captioning architecture is modified to predict aerodynamics problems in combination with an autoencoder. The autoencoder underfits and other architectures are explored for working with small datasets. "A neural network architecture for reduced order modeling of PDEs." Stanford CS221: Artificial Intelligence Class Project. paper poster github
A Novel Application of K-Means Clustering in the Spatial Domain Prior to SVD for Faster ROMs
Spatial clustering is found to be an effective approach for sparsifying and accelerating reduced-order model simulations. "A Spatial Clustering Algorithm for Constructing Local Reduced order Bases for Nonlinear Model Reduction." Stanford CS229: Machine Learning Class Project. link paper github USNCCM Conference Presentation. talk poster talk paper github
Applying Neural Networks to Time-Dependent Aerodynamics using a Video Prediction Model
Aerodynamics solutions were converted into sequences of images for CS231N project. Predictions are more accurate than expected, but the computation is inefficient. "Unsupervised Learning of Time-Dependent CFD Solutions using LSTMs." Stanford CS231N: Convolutional Neural Networks for Visual Recognition Class Project. paper poster github
A Mesh Point Selection Method for Faster and More Accurate ROMs
A mesh point selection method is developed to accelerate ROMs and unexpectedly improves accuracy too. However, it's unknown how to make the point selection computationally efficient for higher dimensional problems. "A reduced order modeling method for improving online computation time and accuracy using mesh coarsening." Stanford AA290: Problems in Aero/Astro Class Project. paper talk github
