PRIMES 2023
Reinforcement Learning Infrastructure for Serverless
PRIMES is a program organized by MIT where high school students work on a year-long research project under the guidance of MIT Ph.D. students
In 2023, my lab-mate Nikita and I co-advised two high school students - Alan Song and Evan Ning.
Alan and Evan worked on using reinforcement learning (RL) for deciding the scale of containers in a serverless environment. They coded up the entire infrastructure needed to run RL-based experiments with serverless. You can find their code here:
With the infrastructure, they obtained preliminary results by applying Deep Q-Learning for horizantal and vertical autoscaling of serverless containers. They present these results in the PRIMES conference. The slides of their presentation are below: