5th Year Masters in Computer Science Thesis Presentation
— 1:30pm
Location:
In Person
-
CMU Swartz Center for Entrepreneurship, 3rd Floor, Tepper Quad
Speaker:
ARI FIORINO
,
Masters StudentComputer Science DepartmentCarnegie Mellon University
https://arifiorino.github.io/index.html
Delayed Gaussian Processes with Time Dependencies and Context
The Gaussian Process (GP) algorithm is able to predict an unknown function from a set of training points. It outputs the mean and uncertainty of its prediction. GP-UCB is a method that uses a GP to optimize any arbitrary function. My thesis is a novel modification of the GP-UCB Algorithm. This modification allows for optimization of an objective over several "episodes" where each episode has a sequence of actions and rewards connected by a time dependency. I present experiments on synthetic data that show that my algorithm performs better than four other state of the art algorithms. Then I ran the algorithm on a COVID dataset to predict a sequence of deaths from a sequence of cases. I present experiments on real world COVID data which shows my algorithm converges to optimal. I also present a very fast GP implementation which runs on a GPU and is only 18 lines long. Thesis Committee: Aarti Singh (Chair) Jeff Schneider Additional Information
For More Information:
tracyf@cs.cmu.edu