ODSIE2023

ODSIE 2023 Keynote Speakers

"Mitigation Strategies of the Occasional Drivers Absenteeism in the Last Mile Delivery"

Prof. Chefi Triki

University of Kent, UK

Short Bio

Chefi Triki is a Senior Lecturer of Operations Research and Logistics Systems at The University of Kent (UK). He holds a Ph.D. in Systems Engineering and Informatics from the University of Calabria (Italy). His major research interests lie in the field of optimization mainly in the context of logistics and resources management. He has published in top scientific journals and served as keynote speaker in a variety of international conferences. His research portfolio includes also several research grants that he led with success in Italy, Oman and Qatar. Dr. Triki has a strong background in developing optimization tools for the network design with application to the transportation procurement, freight distribution, waste collection, groundwater management, tourism planning, etc.

Abstract

Companies can use occasional drivers to increase efficiency on last-mile deliveries. However, as occasional drivers are freelancers without contracts, they can decide at short notice whether they perform delivery requests. If they do not perform their tasks, this is known as driver absenteeism, which obviously disrupts the operations of companies. This talk tackles this problem by developing an auction-based system, including a mitigation strategy to hedge against the absenteeism of occasional drivers. According to this strategy, a driver can bid not only for serving bundles but also to act as a reserved driver. Reserved drivers receive a fee to ensure their presence but are not guaranteed to be assigned to a specific bundle. The problem is modeled as a two-stage stochastic problem with recourse activation. To solve this problem, we develop a self-learning matheuristic. Through an extensive computational study, we show the clear dominance of the newly proposed approach in terms of solution quality, run times, and customers’ perceived quality of service compared against three different deterministic approximations. The Value of the Stochastic Solution, a well-known stochastic parameter is also analyzed. Finally, the identikit of the perfect reserved driver, based on data observed in optimal solutions, is discussed.

"Reinforcement Learning for Solving Operations Research Problems: News Trends and Challenges"

Pr. Safa Bhar LAYEB

LR-OASIS, National Engineering School of Tunis, University of Tunis El Manar, Tunisia

Short Bio

Safa Bhar Layeb is currently a professor of industrial engineering and an active member of the OASIS Laboratory at the National Engineering School of Tunis, Tunisia. Safa holds a Polytechnic Engineering degree, a Master's degree in Mathematical Engineering, a Ph.D. in Applied Mathematics, and a university habilitation (HDR) in Industrial Engineering. She is the founding chair of the African Working Group in Health Systems and member of the Executive committee of the African Federation of Operations Research Societies (AFROS). She has served as guest editor and reviewer for over twenty international journals (ANOR, CAIE, RAIRO,…). She is an active member of the editorial, organizing and scientific committees of several international conferences. She has supervised over twenty Master and Ph.D. thesis in collaboration with international institutes (Politecnico di Milano, Italy; Université de Moncton, Canada; Faculty of Transportation Engineering and Vehicle Engineering, Hongrie; Université de Tlemcen , Algérie; etc.). She has co-authored over thirty papers in reputable indexed journals, fifty papers in international conferences and, twenty chapter books. Her teaching and research interests include Operations research, optimization and data science based-approaches and their applications in healthcare organizations, industrial engineering, logistics, and supply chain management.

Abstract

The emerging trends in data science and artificial intelligence seems representing the future of Operations Research (OR) and decision sciences as a whole. One of the core methods of machine learning is Reinforcement Learning (RL). It basically consists in training a machine learning model in order to make a sequence of decisions based on a trial-and-error approach. Recently, several studies have focused on how classical Operational Research problems are effectively tackled through Reinforcement Learning. Compared with classical metaheuristics and constructive heuristics, RL algorithms have their specific characteristics in solving OR problems. More precisely, RL algorithms reveal the inner structure of the problems by maintaining and improving action value function based on which the scheduling policy is generated and improved. Hence, RL algorithms not only explore a good policy but also use the structure of the problems to adjust the policy, which may explain their potential efficiency for solving large-scale problems. Applications of RL to solve scheduling, routing and loading problems will be presented.

A promising research avenue is to combine Operations Research techniques, precisely heuristics and metaheuristics, with reinforcement learning based-approaches in order to improve the quality of the solutions. New avenues of research on Evolutionary Reinforcement Learning (ERL) and its application for solving OR problems will be described.