Our research is focused on the use of Bayesian statistics and Monte Carlo methods for modern applications. We are interested in developing novel Monte-Carlo related methods for evaluating posterior probability distributions more efficiently compared to state-of-art methods. We use our methods in a wide range of applications including prediction in time series models, policy search in reinforcement learning, and tensor decomposition.
Queueing and Discrete-Event Simulation
Queuing systems are used to model and analyze systems where callers at a call center, production orders at a manufacturing plant, messages in a telecommunication network wait before receiving service/being processed. Together with discrete-event simulation, the research effort can be used to design policies to mitigate the degrading system performance concerning the waiting times.
My research effort is geared to the application of the results from queueing theory for practical problems. One area that I have worked on is modeling production/inventory systems via simple queues. This is how we have shown that production time variability must be lowered, if this is not possible, customers must be quoted a price and lead-time dynamically yet fairly. We explore how a company can distinguish between price sensitive customers and delay sensitive customers while not jeopardizing the fairness principles in serving them. Another area is the call centers, where we show the benefit of offering a callback option for those customers who do not want to wait online. This can be done if a time window can be announced in which a callback customer is reached. We show that albeit agents work more, customer loss can be reduced if such a policy is implemented. At the following site, interested parties can find my research results: