Lila Perkins

Broadly, my research is motivated by the growing computational and energy demands of AI and large-scale scientific computing. More specifically, I develop scalable optimization methods for scheduling computational workloads across distributed computing infrastructure, with the goal of using existing hardware more efficiently and reducing electricity consumption.

The main challenge is solving large-scale integer optimization problems in real time, for example, allocating compute requests across GPU clusters while accounting for hardware characteristics and electricity prices by location. I plan to adapt integer optimization and decomposition techniques from power systems unit commitment to this setting, where similar problems arise in dispatching generators to meet demand. Ultimately, this work aims to support the reliable and efficient integration of large flexible computing loads into a grid increasingly reliant on intermittent renewable generation.