< Sonja Johnson-Yu

Sonja Johnson-Yu


Sonja smiling in a grassy glen


CS PhD Candidate at Harvard

sjohnsonyu@g.[myschool].edu


About

Hello! I'm a rising fourth-year computer science PhD candidate at Harvard, advised by Kanaka Rajan. My aim is to understand animal communication by examining the interplay between signaling and behavior. I'm particularly interested in how animals use signals to achieve their goals and coordinate group behaviors. My current focus is on electrocommunication and electrolocation in weakly electric fish.
Earlier in the PhD, I examined robustness at the intersection of machine learning and optimization, as well as robust planning in multi-objective settings, applied to the problem of wildlife conservation through anti-poaching patrols.

Previously, I received a BA, BS, and MS from Stanford -- my undergrad was a double major in CS and music (concentration in conducting). During my master's in computer science advised by Chris Piech, I applied computer vision and natural language processing to solve problems in sustainability and education.


Recent Work

Characterizing and Improving the Robustness of Predict-Then-Optimize Frameworks.
S. Johnson-Yu, J. Finocchiaro, A. Sinha, K. Wang, Y. Vorobeychik, A. Taneja, & M. Tambe.
GameSec 2023. [PDF]
Modeling Robustness in Decision-Focused Learning as a Stackelberg Game.
S. Johnson-Yu, K. Wang, J. Finocchiaro, A. Taneja, & M. Tambe.
AAMAS 2023 Extended Abstract. [poster] [PDF]
SimGrade: Using Code Similarity Measures for More Accurate Human Grading.
S. Johnson-Yu, N. Bowman, M. Sahami & C. Piech
Educational Data Mining 2021 [poster] [PDF]
ForestNet: Classifying Drivers of Deforestation in Indonesia using Deep Learning on Satellite Imagery.
J. Irvin, H. Sheng, N. Ramachandran, S. Johnson-Yu, S. Zhou, K. Story, R. Rustowicz, C. Elsworth, K. Austin, & A. Ng.
NeurIPS 2020 Workshop on Tackling Climate Change with Machine Learning. [blog post] [preprint]

Collaborations & Real-World Projects

For the first two years of my PhD, I worked on PAWS (Protection Assistant for Wildlife Security), which is a predictive patrolling model that helps rangers to understand areas in global parks that are at high risk for poaching. I was very fortunate to collaborate with SMART and Programme for Belize to work on improving security for Rio Bravo Conservation & Management Area in Belize.