Research

Fundamental Research

I research deep learning optimization techniques and neural net training dynamics, with a focus on ways to characterize better how and what AI models learn. This work aims at the greater goal of finding new ways to train models more efficiently based on provable guarantees, or even preemptively identify training roadblocks that can arise before any actual model has been trained.

Modern deep learning models with billions or trillions of learnable parameters have expanded the capabilities of what complex tasks are possible to learn. However, as model sizes grow towards becoming more heavily overparameterized, training these models becomes a herculean task: requiring immense compute resources, multiple months of training, and potentially hundreds of millions of dollars. My work aims to alleviate these burdens to be able to train models cheaply without jeopardizing downstream performance.

AI for Improving Student Success & Retention

I research novel generative modeling techniques for the early-identification of at-risk students and forecasting adverse outcomes throughout their academic careers. This work aims to better equip human experts to identify at-risk students as early as possible to provide timely interventions and support to improve their academic success.

While timely intervention is the most effective way to prevent student stop-out, traditional institutional processes rely on fragmented, manual reviews that often flag concerns only after a student has undergone significant struggle. My research aims to provide better early-identification of at-risk students, and forecast adverse academic outcomes in a student’s academic career. By reducing the cognitive load on advising teams through automated early-warning systems, this work enables human experts to shift their focus from manual data parsing to delivering personalized, qualitative support.