Near-optimal Control of Complex Authentication Systems
Oğuz Toragay
Department of Industrial and Systems Engineering, Auburn University

Abstract
We consider an authentication system, with several methods available, where requests arrive from users of several classes. Users, such as customers of a secure website, are classified considering their characteristics, such as background, type of request and the priority for the system. Authentication methods can have different capacity and cost. Methods can include checking password, security questions, fingerprint or even a direct phone call from customer service. The decision maker must dynamically assign each incoming request to a method, with the objective of maximizing security and minimizing latency and cost. We model the problem as a multi-class, parallel queueing system and solve it using MDP (Markov Decision Processes). The structure of optimal policies in systems with only one finite-capacity method, is known. We use this structure to propose a new heuristic approach to construct near-optimal policies. To do so, we investigate the behaviour of optimal policy for different users and find upper and lower bounds for the regions in which we know the general form of optimal policy for each user. Using these bounds, we propose a closed-form formula for a heuristic policy. To evaluate the performance of the proposed heuristic method, we generated the necessary random data and solved the model using parallel computing in MATLAB programming environment. Numerical experimentation shows that our approach offers near-optimal performance for a wide range of parameters. Moreover, our heuristic method decreases the necessary computational effort and solving time dramatically. We believe that, using the proposed heuristic can be considered as a practical alternative for solving those systems in which the “curse of dimensionality” plays a major role.
This is joint work with Daniel F. Silva.

Short Bio
Oguz Toragay is a 3 rd year Ph.D. student in the Samuel Ginn College of Engineering at Auburn University where he started his graduate studies in Industrial and Systems Engineering in 2016.
After completing his undergraduate studies at the Khayyam University of Mashhad in Applied Mathematics, Oguz completed his M.Sc. degree at Gazi University in Ankara and then Master of Engineering at Auburn University both in Industrial Engineering. Currently, he continues his research under the supervision of Dr. Daniel Silva. His research interests lie in the area of stochastic optimization, ranging from stochastic OR to the applications of Markov Decision Processes in the queueing systems.
Oguz is the TA and Lab instructor for Manufacturing Systems course in ISE department. Also starting his position in 2017, he has served as the webmaster officer in INFORMS students chapter where he currently continues his service as the secretary officer.
Oguz has dual nationality and knows 4 languages. He plays a traditional music instrument called “Tar”.

Venue
Tuesday, July 3, 2018 at 1.30 pm in IE03