Portfolio

Resume

 

 

AMRL

Microsoft Research (First Author)

 

Submitted a paper, as the first author, to ICLR 2020, as part of a Pre-doc with MSR. (Currently under review and discussion.) The work demonstrates the the sensitivity of modern memory approaches to policy stochasticity and noise in deep RL. We propose a solution framework that improves average return by 19% over a baseline with the same number of parameters and by 9% over a DNC, with many more parameters.

 

 

Paper [Under Review for ICLR 2020]

Stackelberg Vehicles

SDC Lab (Research Member)

 

Worked on social interactions between humans and autonomous vehicles, calculating Stackelberg equilibria in game trees. Paper accepted and to be published at the 2019 International Conference on Social Robotics.

 

Pre-Print Paper

ReNeg

SDC Lab (Research Lead)

 

Leading ongoing research into training an autonomous vehicle via imitation learning with labeled feedback. Used transfer learning with an inception net backbone. Trained the car on the task of lane following and compared the time until the car made a mistake for various architectures and loss functions.

 

Paper [Work in Progress]

Report

Publicity1, Publicity2

Neural Mesh

Research Lead

 

Research into a biologically inspired RNN with a notion of space and conservation of energy, for a graduate AI seminar.

 

arXiv Paper

ACKTR (Implementation)

Group Member

 

Implemented the paper "Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation" by Wu et al. for a graduate seminar.

 

Slides

Report

Food with Friends

Co-founder, designer, and developer

 

Responsible for: multi-threading, algorithms, maps and texting api, user data storage, user management, product ideas, UI, logo, design, dividing work

 

App Store

(screenshots available depending on device)

Paper for Graduate Level Robotics Seminar

Co-author

 

Coded an agent in Minecraft using reinforcement learning and emotion detection implemented via computer vision (OpenCV)

 

Report

For Fun

Home

CaRL

SDC Lab (Group Member)

 

We made a DQN for lane sharing as our first lab project. The goal for the agent was to get to the other side of the truck as fast as possible without crashing. We compared the performance of the RL agent to a human, against three different "dumb" AI opponents that had different aggression levels.

 

Report

GYNN

Group Member

 

Implemented a deep convolutional generative adversarial network (DCGAN) for a deep learning graduate seminar. The neural network completes faces with up to 80% of the pixels missing.

Untitled (Bear/Game)

Project Lead

 

A horror video game featuring Brown University's very own Blueno the Bear – aka Untitled (Lamp/Bear)

 

Please play in full screen!

Untitled (Bear/Game)

Terminal Velocity

Group Member

 

A simple platformer with a twist. (Made in Unity.)

 

Please play in full screen!

Terminal Velocity

LeopardBoy Spaceman

Project Lead

 

Winner of the Kalundborg Game Jam 2017. A simple game with a grappling hook mechanic. Made in Unity in 30 hours minus sleep.

 

LeopardBoy Spaceman