Program
Plenary Talks
28/october
Roger Z. Rios Mercado
Successful Mathematical Optimization Ideas for Territory Design
Roger Z. Rios Mercado
Successful Mathematical Optimization Ideas for Territory Design
Districting or territory design problems involve essentially partitioning decisions. Given a set of basic geographical areas, the idea is to find a partition of these into clusters or districts in such a way that specific planning requirements are met. These requirements depend naturally on the specific application. Districting problems are motivated by very different applications, such as political districting, sales and service territory design, school districting, design of territories for waste collection, to name a few. Through the years many ideas for tackling these very difficult discrete optimization problems have been successfully developed. Given the NP-completeness of districting problems in general, most of the solution methodlogies have relied on heuristic and metaheuristic approachess. However, mathematical programming techniques have also been successul under specific cases. In this talk, we discuss some mathematical programming ideas that successfully been used in exact or decomposition approaches focusing on particular classes of districting problems. We will highlight particular properties that can be intelligently exploited for efficient algorithmic development. Special emphasis will be given to the computational issues and how particular problem structures and properties can be adequately used for efficient algorithmic design. The last part of the talk will include a discussion of current research trends on this fascinating class of discrete optimization problems.
28/october
Ana Paula Cabral Seixas Costa
Advances in multicriteria methods with partial information - methodological innovations, behavioural studies, and the design of decision support systems
Ana Paula Cabral Seixas Costa
Advances in multicriteria methods with partial information - methodological innovations, behavioural studies, and the design of decision support systems
Multicriteria methods that use partial information can be considered an advance in multicriteria decision support, as they require less cognitive effort from decision-makers during the process of eliciting their preferences. These methods differ in the way they require and use partial information, and this includes variations in the amount of structuring and flexibility that the methods permit in the process. The behavioural aspects of decision-makers that influence preference modelling are still little explored, and the decision support systems that support these methods are a critical success factor in providing the decision-maker with a successful experience when using a multicriteria method. In this presentation, we will discuss methodological advances, behavioural studies, and how these advances and studies have contributed to modulating methods and improving the design of decision support systems.
29/october
Dylan F. Jones
Multiple Objective Applications in Renewable Energy Planning and Arctic Innovation Need Selection
Dylan F. Jones
Multiple Objective Applications in Renewable Energy Planning and Arctic Innovation Need Selection
This talk presents three multiple objective applications. These originate from, or are inspired by, collaborative work on projects funded under the EU Framework 7 and Horizon programmes (leanwind.eu, arcsar.eu, ai-arc.eu). Two case studies of the usage of goal programming to aid decisions in the transition to renewable energy are given. The first is a model for the selection of marine renewable energy sites in the United Kingdom, considering a balance between geographical clusters and different energy types. The usage of interval coefficient goal programming to handle uncertainty over a set of economic, environmental and social sustainability criteria is described. A second renewable energy case study dealing with the renewable energy transition in China is presented. A desired energy configuration in 2050 under different national scenarios is generated using extended network goal programming (ENGP). The balance between national level, different regions and a set of sustainability criteria is optimised. A multi-period stochastic generation expansion planning model is then formulated and solved in order to give the optimal temporal decisions in each region in order to achieve the 2050 optimal configuration given by the ENGP model. The final case study involves the selection of a priority set of Arctic Innovation Needs for safety and security in the Arctic and North Atlantic. Collaboration with search and research (SAR) and Marine Environmental Response (MER) practitioners is first described. The resulting need classification scheme and its assessment by importance and difficultly is then detailed. A balanced knapsack goal programming model and its usage in deriving a priority set of innovation needs are shown. Some results in terms of SAR and MER policy and practice are detailed. Finally, conclusions as to future applications of multiple criteria optimisation techniques are drawn.
30/october
Susana Mondschein
Optimizing Cancer Prevention and Control Through Operations Management and Data Science
Susana Mondschein
Optimizing Cancer Prevention and Control Through Operations Management and Data Science
In healthcare, operations management and data science present substantial opportunities to improve the efficiency and effectiveness of cancer prevention and control programs. This presentation explores various aspects of these programs, addressing critical challenges such as limited financial and medical resources, diverse risk groups, and the distinctions between individual and population-level decision-making. Key topics to be covered include:
management and data science in advancing cancer control strategies and addressing the various
challenges faced in this critical area of healthcare.
- The Importance of Data Quality: We will explore the necessity for high-quality, accessible, comprehensive, and up-to-date data in shaping effective decision-making processes.
- Addressing Inequalities in Cancer Survival: Analyzing how disparities in cancer outcomes impact different populations and identifying strategies to mitigate these inequalities.
- Complexity of Screening Optimization Models: Discussing the mathematical challenges of developing cancer screening models.
- Risk Assessment: Examining the role of machine learning models in enhancing cancer risk assessment and their application in improving preventive measures.
management and data science in advancing cancer control strategies and addressing the various
challenges faced in this critical area of healthcare.
31/october
Thibaut Vidal
OR with a White Hat : Evidencing Privacy Vulnerabilities in ML Models
Thibaut Vidal
OR with a White Hat : Evidencing Privacy Vulnerabilities in ML Models
The use of personal data and the integration of decision-support systems and machine learning algorithms in finance, medicine, and many other domains can profoundly impact human lives. Consequently, extensive efforts have been made to improve machine learning pipelines, e.g., making them more accurate, interpretable, and secure. In this talk, we discuss the synergy between combinatorial optimization and machine learning. We focus primarily on tree ensembles (including random forests and gradient boosting), a popular family of models with good empirical performance which is often used as a more transparent replacement to neural networks. We delve into the application of mathematical optimization to various critical tasks in trustworthy machine learning. Specifically, we present an optimization-based reconstruction attack capable of effectively recovering a dataset used to train a random forest model. By framing the dataset reconstruction problem as a combinatorial optimization problem with a maximum likelihood objective and leveraging constraint programming to solve it, our experiments demonstrate that random forests are highly susceptible to complete dataset reconstruction, even with very few trees. These findings highlight a significant vulnerability in commonly used ensemble methods, emphasizing the need for attention and mitigation.
1/november
Tava Olsen
Paying More to Get it Faster: What, When, and How
Tava Olsen
Paying More to Get it Faster: What, When, and How
This talk gives an overview of my and others research in leadtime-based pricing. We consider the what, when, and how of paying more to get a product or service faster, particularly in capacity constrained settings. Canonical applications include custom furniture and DNA testing. At the time of ordering customers may be presented with a menu of leadtimes and prices, just as companies like Amazon do for delivery times. However, due to the provider’s capacity constraints these leadtimes need to reflect what can actually be produced (and when). Most of the talk considers a make-to-order manufacturer or service provider whose customers are time sensitive. Particular aspects considered are: non-linear waiting cost curves, heterogeneous customers, and quotes that affect customer satisfaction. The methodology of the work consists of various forms of stochastic modelling. However, there has also been some behavioural experiments to get a better sense of the correct form for the cost of waiting relative to what is quoted, and this will also be presented briefly.
