Adarsh Mital's repositories
6787_project
CS6780 “Advanced Machine Learning”. Implemented multiple Federated Learning averaging methods in a Differentially Private setting and measured relative impact on model accuracy and fairness. Worked jointly with Caleb Berman of the Cornell MPS program
AIEthicsResearch
Capstone report for INFO4301 “Ethics in New Media”. Surveyed emerging ability of transformer-based AI models to hold massive context. Identified early work in using this ability to predict human behavior, and projected how this might progress when sensory signals and biomarkers are combined with overt words and actions
IndependentResearch
Report for INFO4900 Independent Research under Prof. Dawn Schrader. Surveyed bias detection and mitigation methods in language models. Identified emerging Language Model tasks where existing mechanisms fail. Designed a novel fairness test and proposed a framework to update large language models when what society considers fair changes.
ORIE_4741_Project
ORIE4741 “Learning with Big Messy Data”. Implemented sentiment and mood (Happy, Angry, Sad, and Relaxed) classification of song lyrics that are less structured and logical, with pauses and emphasis important to the meaning. Models trained on flowing prose don’t directly work. Joint work with Aliyah Geer and Elise Kronbichler