Programme
Tuesday 25 November
- 09:00 Registration
- 10:45 Opening session
- 11:00 Plenary session 1. Kenneth Sörensen
- 12:00 Lunch in the cafeteria
- 12:45 Parallel sessions TU-A
- 14:15 Short break
- 14:30 Parallel sessions TU-B
- 16:00 Coffee break and refreshments
- 16:30 Parallel sessions TU-C (until 18:00)
- 19:00 Welcome reception in the main hall
- 19:30 Conference dinner (until 22:30)
Keynote:
Metaheuristics – Towards an Empirical Science
Professor Kenneth Sörensen, University of Antwerp, Belgium
"Metaheuristics – Towards an Empirical Science"
There is something rotten in the field of metaheuristics. “Novel” algorithms based on increasingly absurd metaphors continue to flood our journals. Obvious frauds rise to the top of citation rankings.
Could this be related to how we conduct research? Could our 'horse race' mentality—where algorithmic performance alone determines scientific merit—actually be undermining the intellectual foundation of the field? And could the effort to ground heuristics in mathematics have been doomed from the start?
This talk is a call to let go of the idea that the only scientific way to study (meta)heuristics is through mathematics. Rather, we propose to transform the field into an empirical science.
In Turkeš et al. (2021), we published the first meta-analysis in the field. Using techniques from medical research, we examined the effect of adaptiveness in ALNS and found it significantly smaller than widely believed. At the same time, our research also demonstrated that widely accepted beliefs in the field can rest on not much more than anecdote, and that mathematical theory has not filled that gap. We hoped this work would inspire similar efforts, but those did not materialize.
Much of the necessary data—algorithms, instances, results—already exists, but many changes are still needed to make empirical research possible at scale. In Swan et al. (2022), we proposed several of such practical changes: standardized protocols, reusable algorithm components, white-box problem descriptions, open repositories, and remote execution platforms.
What is needed first, though, is a shift in mindset: away from the algorithmic performance horse race and towards a common research effort aimed at discovering true knowledge.
Wednesday, 26 November
- 08:30 Parallel sessions WE-A
- 10:00 Coffee break and small refreshments
- 10:30 Parallel sessions WE-B
- 12:00 Short break
- 12:15 Plenary session 2 - Lars Magnus Hvattum
- 13:15 Closing by conference chair. Lunch outside conference rooms
Keynote: Operations research as support for football clubs
Lars Magnus Hvattum, Professor in Quantitative Logistics Molde University College, Norway
"Operations research as support for football clubs"
Most football clubs are deeply rooted in traditions, and make decisions based solely on human judgement and experience. However, it is possible to detect a recent trend that clubs are collecting ever more data and some of them try to exploit these data to make better decisions. This talk addresses some areas where sports analytics and operations research can help football clubs to gain insights that may otherwise be missed.
We start by discussing rating systems for teams, and how these can be used to gain insights into the relative strength of different teams. As an example of how this can be used to gain additional insights, we look at how the introduction of artificial turf impacted the sporting performance of the home teams involved.
Next, we look at how one can build rating models for individual football players. Using different types of data will lead to different types of models, with different properties. We focus on simple match sheet data, where a popular model type leads to a formulation of an unconstrained quadratic program of a very large size. The size of the problem makes gradient descent an attractive solution method. The resulting ratings can be used in the early scouting process of a professional club.
However, the ratings can also be used in prediction models for match outcomes, which in turn can be used to predict the outcomes of competitions, ranging from double round robin leagues to EURO play-offs. Finally, we discuss how binary integer programming models can be used to support building a squad of players while considering both financial concerns and sporting performance.