Data Science Research Center Seminar | An approximate grid-based solution method for POMDPs | The case for personalized cervical cancer screening policies
Join us for a talk from Raha Akhavan-Tabatabaei, Sabanci Business School, Turkey.
Partially Observable Markov Decision Process (POMDP) model, an analytical framework for optimizing theory, can be applied in making the screening decision: the type of screening test, the individual cancer risk, and the test characteristics. Age-specific unobservable disease progression and regression rates are also reflected in the health state transition probabilities of the underlying Markov chain, resulting in six partially observable states. Given the large size of POMDPs, they become computationally intensive if not intractable. Hence, we propose to solve the problem by applying the approximate grid-based solution method and use Gaussian Process Regression (GPR) to enquire value functions at non-grid belief states. We provide insight on the health outcomes of implementing our patient-tailored screening policy.
Raha Akhavan-Tabatabaei is an associate professor of operations management and program director of the M.Sc. in business analytics at Sabanci Business School in Istanbul, Turkey. Prior to this role, she held positions of associate professor of operations research and founding director of the master’s in analytics at Universidad de los Andes in Bogota, Colombia, and senior industrial engineer at Intel Corp. in Arizona, U.S. She has a Ph.D. in industrial engineering and operations research from North Carolina State University. Her research focus is stochastic modeling and data-driven decision-making with applications in health care, logistics, revenue management and reliability, among others.
The talk will be in English and is open only to members of DKU community.
Register using this link: https://duke.qualtrics.com/jfe/form/SV_0J1NYioN36bI0cZ
For questions, contact firstname.lastname@example.org.