Summary
This semester programme aims to combine optimization and machine learning, in order to develop data-centric methods attaining both performance guarantees and explainability. This includes the design of algorithms incorporating machine-learned predictions of uncertain quality, methods using machine-learned models within algorithms, using machine-learning to model practical problems and using optimization techniques to analyze and improve machine learning algorithms. The programme consists of a bootcamp, a PhD School, two satellite workshops on different aspects of the topic and a seminar++ series. The organizational team comprises diverse international experts in optimization and machine learning, featuring both junior and senior researchers.
Introduction
Traditional optimization algorithms are based on abstract models that attempt to capture the essential properties of a practical problem. Such algorithms typically provide strong guarantees by assessing their performance on worst-case instances under this model, but which can be overly pessimistic with respect to their actual performance on practical instances. On the other hand, machine learning methods are trained and analyzed on practical data but generally offer no guarantee and poor explainability.
The scientific challenge addressed in this Research Semester Programme “Learning-Enhanced Optimization” is to combine optimization and machine learning to develop data-centric methods attaining both performance guarantees and explainability. This includes the design of algorithms incorporating machine-learned predictions of uncertain quality, methods using machine-learned models within an optimization algorithm, using machine-learning to model practical problems and using optimization techniques to analyze and improve algorithms used in machine learning.
Preliminary programme
A tentative overview of the events and their respective dates is given below:
- Bootcamp: Interplay Between Machine Learning and Optimization
Duration: 1 day, early September 2025, tentative date: Sep. 1, 2025
Goal: kick-off event highlighting the different aspects of the semester programme - International PhD School: Machine Learning and Optimization
Duration: 4 days, early September 2025, tentative dates: Sept. 2-5, 2025
Goal: 3-4 invited lecturers offering lectures/tutorials for PhD students, with the purpose of training and community building - Satellite Workshop 1: Learning Augmented Algorithms
Duration: 3-4 days, early October 2025, tentative dates: Oct. 7-10, 2025
Goal: get experts of the field together (both national and international), combination of survey talks (keynotes), specialized talks (contributed), and open problem sessions. - Satellite Workshop 2: Learned Methods for Operations Research
Duration: 3-4 days, early November 2025, tentative dates: Nov. 3-6, 2025)
Goal: get experts of the field together (both national and international), combination of survey talks (keynotes), specialized talks (contributed), and open problem sessions. - Seminar++ Series: Advances in Fusing Optimization and Learning
Duration: half-day events, biweekly, dates TBD
Goal: open but with strong focus on getting Dutch community together, contributed talks, lightning talks by students, open problem plus collaboration session
The events are open to both national and international researchers. Students are especially welcome. The respective calls will be sent out in due time.
