Hello! I'm Sky.
With a focus on natural language processing, my research aims to make advances on creating more efficient, more effective, and more robust linguistic communication policies, with a long-term goal of bulding smarter, more personalized agents.
My current research projects focus on situated human-agent communication and computational sociolinguistics.
(@EAGLe Lab) I'm interested in looking at how to build agents 🤖 that think 🤔 like humansTM, communicate with humans, and that are able to learn efficiently from humans and other agents. In this regard, grounded, situated language learning and machine theory of mind are areas of particular interest to me. How do humans communicate with each other in coorperative or adversarial scenarios? In times of uncertainty? How is communication predictive of actions, and can we incorporate these ideas in building better, more efficient agents?
(@Blablablab) Looking at how language 🤬 itself is reflective of society's beliefs and values, and how such norms have evolved over time is, too, an area of intrigue. What causes these shifts in cultural norms, who 👪 are the most affected, and how can we best capture these changes? Here, computational sociolinguistics comes into play. In particular, I'm looking at how & where social perceptions of gender have evolved over time, the disparities that exist, and their sources of change.
(@SRG) Previously, I focused on adversarial robustness and reinforcement learning. There, I looked at creating more robust 💪 and more modular defenses against adversarial attacks, and on creating more sample efficient ⏩ learning methods for agents with combinatorial action spaces.