We examine a directed network on which sensors exist at a subset of the nodes. At each node, a ta... more We examine a directed network on which sensors exist at a subset of the nodes. At each node, a target (e.g., an intruder in a physical network, a fire in a building, or a virus in a computer network) may or may not exist. Sensors detect the presence of these targets by monitoring the nodes at which they are located and all nodes adjacent from their positions. However, in practical settings a limited-cardinality subset of sensors might fail. A failed sensor may report false positives or negatives, and as a result, the network owner may not be able to ascertain whether or not targets exist at some nodes. If it is not possible to deduce whether or not a target exists at a node with a given set of sensor readings, then the node is said to be ambiguous. We show that a network owner must solve a series of combinatorial optimization problems to determine which nodes are ambiguous. Furthermore,
we determine the worst-case number of ambiguous nodes by optimizing over the set of all sensor readings that could possibly arise. We also present mathematical programming formulations for these problems under varying assumptions on how sensors fail, and on what assumptions a network owner makes on how sensors fail. Our computational results illustrate how these varying assumptions impact the number of ambiguous nodes.
This chapter provides an introductory analysis of linear programming foundations and large-scale ... more This chapter provides an introductory analysis of linear programming foundations and large-scale methods. The chapter begins by discussing the basics of linear programming modeling and solution properties, duality principles for linear programming problems, and extensions to integer programming methods. We then develop Benders decomposition, Dantzig-Wolfe decomposition, and Lagrangian optimization procedures in the context of network design and routing problems that arise in telecommunications operations research studies. The chapter closes with a brief discussion and list of basic references for other large-scale optimization algorithms that are commonly used to optimize telecommunications systems, including basis partitioning, interior point, and heuristic methods.
• Our MDP model for the courier assignment task characterizes on-demand meal delivery service. • ... more • Our MDP model for the courier assignment task characterizes on-demand meal delivery service. • We tailor deep reinforcement learning algorithms to address the problem in a dynamic environment. • We incorporate the notion of order rejection to reduce the number of late orders. • We investigate the importance of intelligent repositioning of the couriers during their idle times.
This paper proposes a real-life application of deep reinforcement learning to address the order d... more This paper proposes a real-life application of deep reinforcement learning to address the order dispatching problem of a Turkish ultra-fast delivery company, Getir. Before applying off-the-shelf reinforcement learning methods, we define the specific problem at Getir and one of the solutions the company has implemented. We discuss the novel aspects of Getir’s problem compared to the state-of-the-art order dispatching studies and highlight the limitations of Getir’s solution. The overall aim of the company is to deliver to as many customers as possible within 10 minutes. The orders arrive throughout the day, and centralized warehouses in the regions decide whether an incoming order should be served or canceled depending on their couriers’ shifts and status. We use Deep Q-networks to learn the actions of warehouses, i.e., accepting or canceling an order, directly from state dimensions using reinforcement learning. We design the networks with two different rewards. We conduct empirical analyses using real-life data provided by Getir to generate training samples and to assess the models’ performance during a selected 30-day period with a total of 9880 orders. The results indicate that our proposed models are able to generate policies that outperform the rule-based heuristic employed in practice.
International Series in Operations Research & Management Science, 2010
Abstract This chapter provides an introductory analysis of linear programming foundations and lar... more Abstract This chapter provides an introductory analysis of linear programming foundations and large-scale methods. The chapter begins by discussing the basics of linear programming modeling and solution properties, duality principles for linear programming problems, and ...
Springer Proceedings in Mathematics & Statistics, 2013
We examine a directed network on which sensors exist at a subset of the nodes. At each node, a ta... more We examine a directed network on which sensors exist at a subset of the nodes. At each node, a target (e.g., an intruder in a physical network, a fire in a building, or a virus in a computer network) may or may not exist. Sensors detect the presence of these targets by monitoring the nodes at which they are located and all nodes adjacent from their positions. However, in practical settings a limited-cardinality subset of sensors might fail. A failed sensor may report false positives or negatives and, as a result, the network owner may not be able to ascertain whether or not targets exist at some nodes. If it is not possible to deduce whether or not a target exists at a node with a given set of sensor readings, then the node is said to be ambiguous. We show that a network owner must solve a series of combinatorial optimization problems to determine which nodes are ambiguous. Furthermore, we determine the worst-case number of ambiguous nodes by optimizing over the set of all sensor readings that could possibly arise. We also present mathematical programming formulations for these problems under varying assumptions on how sensors fail, and on what assumptions a network owner makes on how sensors fail. Our computational results illustrate how these varying assumptions impact the number of ambiguous nodes.
Abstract Given an undirected graph, a bramble is a set of connected subgraphs (called bramble ele... more Abstract Given an undirected graph, a bramble is a set of connected subgraphs (called bramble elements) such that every pair of subgraphs either contains a common node, or such that an edge (i, j) exists with node i belonging to one subgraph and node j belonging ...
Robust and cost-effective distribution is critical to any home delivery network growing company, ... more Robust and cost-effective distribution is critical to any home delivery network growing company, both to meet demand under normal conditions and to adapt to temporary disruptions. Home healthcare is anticipated to be a rapidly growing modality of healthcare, itself the largest industry in the US and rife with optimization needs in areas such as logistics, scheduling, and supply chains. We develop two mixed integer programming models to optimize forward storage locations in the supply chain of a national consumable medical supplies company with consistent monthly repeating demand, temporary disruption of facility operations, and remote international manufacturers. Modified p-median single and multi-echelon models are used to determine optimal locations of warehouses and distribution facilities that minimize total transportation cost, with 13% savings in one application (approximately $1.4 million annually). Sensitivity analyses to a range of scenarios suggest that the optimal solution is robust across a number of potential scenarios.
Given an undirected graph, a bramble is a set of connected subgraphs (called bramble elements) su... more Given an undirected graph, a bramble is a set of connected subgraphs (called bramble elements) such that every pair of subgraphs either contains a common node, or such that an edge (i,j) exists with node i belonging to one subgraph and node j belonging to the other. In this paper, we examine the problem of nding the bramble number of a graph, along with a set of bramble elements that yields this number. The bramble number is the largest cardinality of a minimum hitting set over all bramble elements on this graph. A graph with bramble bramble number k has a treewidth of k - 1. We provide a branch-and-price-and-cut method that generates columns corresponding
to bramble elements, and rows corresponding to hitting sets. We then examine the computational efficacy of our algorithm on a randomly generated data set.
We examine a directed network on which sensors exist at a subset of the nodes. At each node, a ta... more We examine a directed network on which sensors exist at a subset of the nodes. At each node, a target (e.g., an intruder in a physical network, a fire in a building, or a virus in a computer network) may or may not exist. Sensors detect the presence of these targets by monitoring the nodes at which they are located and all nodes adjacent from their positions. However, in practical settings a limited-cardinality subset of sensors might fail. A failed sensor may report false positives or negatives, and as a result, the network owner may not be able to ascertain whether or not targets exist at some nodes. If it is not possible to deduce whether or not a target exists at a node with a given set of sensor readings, then the node is said to be ambiguous. We show that a network owner must solve a series of combinatorial optimization problems to determine which nodes are ambiguous. Furthermore,
we determine the worst-case number of ambiguous nodes by optimizing over the set of all sensor readings that could possibly arise. We also present mathematical programming formulations for these problems under varying assumptions on how sensors fail, and on what assumptions a network owner makes on how sensors fail. Our computational results illustrate how these varying assumptions impact the number of ambiguous nodes.
This chapter provides an introductory analysis of linear programming foundations and large-scale ... more This chapter provides an introductory analysis of linear programming foundations and large-scale methods. The chapter begins by discussing the basics of linear programming modeling and solution properties, duality principles for linear programming problems, and extensions to integer programming methods. We then develop Benders decomposition, Dantzig-Wolfe decomposition, and Lagrangian optimization procedures in the context of network design and routing problems that arise in telecommunications operations research studies. The chapter closes with a brief discussion and list of basic references for other large-scale optimization algorithms that are commonly used to optimize telecommunications systems, including basis partitioning, interior point, and heuristic methods.
• Our MDP model for the courier assignment task characterizes on-demand meal delivery service. • ... more • Our MDP model for the courier assignment task characterizes on-demand meal delivery service. • We tailor deep reinforcement learning algorithms to address the problem in a dynamic environment. • We incorporate the notion of order rejection to reduce the number of late orders. • We investigate the importance of intelligent repositioning of the couriers during their idle times.
This paper proposes a real-life application of deep reinforcement learning to address the order d... more This paper proposes a real-life application of deep reinforcement learning to address the order dispatching problem of a Turkish ultra-fast delivery company, Getir. Before applying off-the-shelf reinforcement learning methods, we define the specific problem at Getir and one of the solutions the company has implemented. We discuss the novel aspects of Getir’s problem compared to the state-of-the-art order dispatching studies and highlight the limitations of Getir’s solution. The overall aim of the company is to deliver to as many customers as possible within 10 minutes. The orders arrive throughout the day, and centralized warehouses in the regions decide whether an incoming order should be served or canceled depending on their couriers’ shifts and status. We use Deep Q-networks to learn the actions of warehouses, i.e., accepting or canceling an order, directly from state dimensions using reinforcement learning. We design the networks with two different rewards. We conduct empirical analyses using real-life data provided by Getir to generate training samples and to assess the models’ performance during a selected 30-day period with a total of 9880 orders. The results indicate that our proposed models are able to generate policies that outperform the rule-based heuristic employed in practice.
International Series in Operations Research & Management Science, 2010
Abstract This chapter provides an introductory analysis of linear programming foundations and lar... more Abstract This chapter provides an introductory analysis of linear programming foundations and large-scale methods. The chapter begins by discussing the basics of linear programming modeling and solution properties, duality principles for linear programming problems, and ...
Springer Proceedings in Mathematics & Statistics, 2013
We examine a directed network on which sensors exist at a subset of the nodes. At each node, a ta... more We examine a directed network on which sensors exist at a subset of the nodes. At each node, a target (e.g., an intruder in a physical network, a fire in a building, or a virus in a computer network) may or may not exist. Sensors detect the presence of these targets by monitoring the nodes at which they are located and all nodes adjacent from their positions. However, in practical settings a limited-cardinality subset of sensors might fail. A failed sensor may report false positives or negatives and, as a result, the network owner may not be able to ascertain whether or not targets exist at some nodes. If it is not possible to deduce whether or not a target exists at a node with a given set of sensor readings, then the node is said to be ambiguous. We show that a network owner must solve a series of combinatorial optimization problems to determine which nodes are ambiguous. Furthermore, we determine the worst-case number of ambiguous nodes by optimizing over the set of all sensor readings that could possibly arise. We also present mathematical programming formulations for these problems under varying assumptions on how sensors fail, and on what assumptions a network owner makes on how sensors fail. Our computational results illustrate how these varying assumptions impact the number of ambiguous nodes.
Abstract Given an undirected graph, a bramble is a set of connected subgraphs (called bramble ele... more Abstract Given an undirected graph, a bramble is a set of connected subgraphs (called bramble elements) such that every pair of subgraphs either contains a common node, or such that an edge (i, j) exists with node i belonging to one subgraph and node j belonging ...
Robust and cost-effective distribution is critical to any home delivery network growing company, ... more Robust and cost-effective distribution is critical to any home delivery network growing company, both to meet demand under normal conditions and to adapt to temporary disruptions. Home healthcare is anticipated to be a rapidly growing modality of healthcare, itself the largest industry in the US and rife with optimization needs in areas such as logistics, scheduling, and supply chains. We develop two mixed integer programming models to optimize forward storage locations in the supply chain of a national consumable medical supplies company with consistent monthly repeating demand, temporary disruption of facility operations, and remote international manufacturers. Modified p-median single and multi-echelon models are used to determine optimal locations of warehouses and distribution facilities that minimize total transportation cost, with 13% savings in one application (approximately $1.4 million annually). Sensitivity analyses to a range of scenarios suggest that the optimal solution is robust across a number of potential scenarios.
Given an undirected graph, a bramble is a set of connected subgraphs (called bramble elements) su... more Given an undirected graph, a bramble is a set of connected subgraphs (called bramble elements) such that every pair of subgraphs either contains a common node, or such that an edge (i,j) exists with node i belonging to one subgraph and node j belonging to the other. In this paper, we examine the problem of nding the bramble number of a graph, along with a set of bramble elements that yields this number. The bramble number is the largest cardinality of a minimum hitting set over all bramble elements on this graph. A graph with bramble bramble number k has a treewidth of k - 1. We provide a branch-and-price-and-cut method that generates columns corresponding
to bramble elements, and rows corresponding to hitting sets. We then examine the computational efficacy of our algorithm on a randomly generated data set.
Uploads
Books by Sibel B Sonuc
we determine the worst-case number of ambiguous nodes by optimizing over the set of all sensor readings that could possibly arise. We also present mathematical programming formulations for these problems under varying assumptions on how sensors fail, and on what assumptions a network owner makes on how sensors fail. Our computational results illustrate how these varying assumptions impact the number of ambiguous nodes.
Papers by Sibel B Sonuc
to bramble elements, and rows corresponding to hitting sets. We then examine the computational efficacy of our algorithm on a randomly generated data set.
Talks by Sibel B Sonuc
we determine the worst-case number of ambiguous nodes by optimizing over the set of all sensor readings that could possibly arise. We also present mathematical programming formulations for these problems under varying assumptions on how sensors fail, and on what assumptions a network owner makes on how sensors fail. Our computational results illustrate how these varying assumptions impact the number of ambiguous nodes.
to bramble elements, and rows corresponding to hitting sets. We then examine the computational efficacy of our algorithm on a randomly generated data set.