4 présentations

• 
  15h30 - 15h55

  Data-driven decision making under uncertainty with entropic risk measure

  • Utsav Sadana, prés., University of Montreal
  • Erick Delage, GERAD, HEC Montréal
  • Angelos Georghiou, University of Cyprus

  Entropic risk measure is widely used to account for tail risks associated with an uncertain parameter. We study distributionally robust optimization problems with entropic risk measures and use cross-validation to tune the radius (hyperparameter) of the ambiguity set. However, with limited data, the empirical average of the entropic risk associated with scenarios in the validation set underestimates the true entropic risk. We provide two procedures that first learn Gaussian mixture models and then use bootstrapping to identify scaling parameters to correct the bias in the entropic risk estimated using the validation data. We show that high losses could be avoided using our debiasing procedure in project selection and portfolio optimization problems.

• 
  15h55 - 16h20

  An improved Self-Adaptive Hybrid Approach based on History-Driven Methods

  • Sina Alizadeh, prés., University of Regina
  • Malek Mouhoub, University of Regina

  We propose an improved self-adaptive hybrid approach based on history-driven methods called History-driven Particle Swarm Optimization-Simulated Annealing (HdPSO-SA) . Collaboration is performed using a Self-Adaptive Binary Space Partitioning tree (SABSP tree) to partition search space and guide the hybrid framework to the most promising sub-region of a given continuous problem to solve. The proposed hybrid framework includes of three phases. In the first phase, a Self-Adaptive Binary Space Partitioning tree (SABSP tree) is applied in PSO to gather essential information, create the fitness values landscape, and partition the search space. In the second phase, an intelligent controller learns the SA-BSP maturity condition to switch between exploration (HdPSO) and exploitation (SA). The third phase limits the search space to the most promising sub-region. Then, the information on the best solution (fitness value and position) will be given to SA to exploit the limited search space. The proposed HdPSO-SA is compared to several metaheuristics on ten well-known unimodal and multimodal continuous optimization benchmarks. The results demonstrate the superiority of HdPSO-SA in returning a good quality solution in an efficient running time.

• 
  16h20 - 16h45

  Solving an Optimal Search Path Problem Considering Search Area Selection using GRASP and Benders Decomposition Algorithms

  • Saeid Abbasiparizi, prés., Ph.D. Student at the Department of Operations and Decision Systems, Université Laval, Québec, Canada
  • Michael Morin, Assistant Professor at the Department of Operations and Decision Systems, Université Laval, Québec, Canada
  • Irène Abi-Zeid, Full Professor at the Department of Operations and Decision Systems, Université Laval, Québec, Canada

  In Canada, a significant number of search and rescue incidents occur annually, some of which require the deployment of major search operations. To support the planning process of search operations, we address the problem of selecting the search rectangles to be visited by a search and rescue unit along with their order of visit. The objective is to maximize the probability of finding the search object. When the search area is large, many search rectangles must be considered, leading to higher complexity due to numerous binary variables. We propose to address this challenge with rectangles filtering, a Greedy randomized adaptive search procedure (GRASP) and Benders decomposition. Our results show the efficacy of the proposed solution approach on realistic case studies.

• 
  16h45 - 17h10

  An imitation-based learning approach using DAgger for the Casual Employee Call Timing Problem

  • Prakash Gawas, prés., Polytechnique Montreal

  Predictive models are increasingly important in enhancing decision-making processes. This study proposes an innovative approach utilizing DAgger, an imitation learning algorithm, to iteratively train a policy for addressing stochastic sequential decision problems. These problems can be challenging, especially when expert input is costly or unavailable. Our focus lies in crafting an effective expert within the DAgger framework, drawing from deterministic solutions derived from contextual scenarios generated at each decision point. Subsequently, a predictive model is developed to mimic the expert’s behavior, aiding real-time decision-making. To illustrate the applicability of this methodology, we address a dynamic employee call-timing issue concerning the scheduling of casual personnel for on-call work shifts. The key decision involves determining the optimal time to contact the next employee in seniority order, allowing them to select a preferred shift. Uncertainty arises from the varying response times of employees. The goal is to strike a balance between minimizing schedule changes induced by early notifications or calls and avoiding unassigned shifts due to late notifications. Unlike traditional predict-and-optimize approaches, our method utilizes optimization to train learning models that establish connections between the system’s current state and the expert’s wait time. We apply our algorithm using data provided by our industrial partner to derive an operational policy. Results demonstrate the superiority of this policy over the current heuristic method in use.