Projects
Some of the projects that I've worked on
Multi-Task Reinforcement Learning: Single-Agent to Multi-Agent Systems
Thesis work on multi-task reinforcement learning. Used novel multi-task incremental learning algorithmIL-SOAR (Incremental Learning with Second-Order Approximation Regularization) to mitigate some of the effects of catastrophic forgetting.
DARPA Gamebreaker
Used machine learning to quantitatively assess game balance in Starcraft II to identify parameters which control balance. Developed and analyzed massive datasets for machine learning using SC2, and programmed an interactive dashboard. Trained a model with accurately predicted the win probabilities using only army compositions.
DARPA ADT
Developed AI fighter jet technology that beat out other competitors and an elite Air Force pilot. Contributed novel agents and reward schemes for training our agent in a RL league. Winner of the DARPA ADT competition.
The Clicker Band
Developed a wearable assistive technology device which allowed helped a student with disability use his computer. Winner of the SourceAmerica Design Challenge with ~$5000 in prize money.