Kirthevasan is a fourth year Ph.D. student working with Jeff Schneider and Barnabás Póczos in the Machine Learning Department at Carnegie Mellon University. He completed his B.Sc in Electronic and Telecommunication Engineering at the University of Moratuwa, Sri Lanka. His primary research interests are in bandit optimisation and other sequential decision making problems but he also works on various topics in nonparametric statistics.

Research Summary

Many scientific and engineering tasks can be cast as bandit optimisation problems, where we need to sequentially evaluate a noisy black box function with the goal of finding its optimum. Typically, each function evaluation incurs a large computational or economic cost, and we need to keep the number of evaluations to a minimum. Some applications include tuning hyper-parameters of statistical models, scientific studies, industrial design and on-line advertising. Today, there is an increasing need to scale up conventional bandit methods which work reliably only in small scale problems. Some modern challenges include extremely expensive evaluations and the need to optimise in high dimensional spaces. At the same time, we now have new opportunities to conduct multiple evaluations in parallel and cheaply approximate the function of interest. Kirthevasan’s research focuses on developing theoretically grounded methods to address such challenges in order to meet emerging demands in large scale bandit applications. He works on a mixture of both theory and applications, with the latter focusing on hyper-parameter tuning and computational astrophysics.