For best experience please turn on javascript and use a modern browser!
You are using a browser that is no longer supported by Microsoft. Please upgrade your browser. The site may not present itself correctly if you continue browsing.
The research project Optimization for and with Machine Learning (OPTIMAL) is an ENW-GROOT research project funded by NWO. It is a collaboration between researchers from Centrum Wiskunde & Informatica (CWI), Delft University of Technology (TU Delft) and the University of Amsterdam (UvA). It started in 2020 and will end in 2025.

Focus and scope

A key component of machine learning is mathematical optimisation, that is used, for example, to train neural networks. The goal of this project is to provide new analysis and tools for optimisation problems and algorithms arising in machine learning, but also to use insights and tools from machine learning to improve optimisation methods. This explains the project title Optimization for and with Machine Learning (OPTIMAL).

Project management

The project consists of four connected work packages. The first two work packages are related to 'optimisation for machine learning’. In the first workpackage we will investigate why the optimisation methods currently used in machine learning are often successful in practice and analyse the limits of their computational tractability. The second workpackage is aimed at enhancing the existing optimisation algorithms and developing new ones to obtain more accurate machine learning models in an efficient way.

The last two workpackages are related to ‘optimisation with machine learning’. The third work package is aimed at using machine learning to obtain data-centric approximation and optimisation algorithms. We will develop algorithms that adapt to the specific data characteristics of the problem instance. The advantage of such data-centric algorithms is more accurate solutions and/or less computation time. In the fourth workpackage we will develop a data-centric optimisation modeling approach. In such an approach parts of the resulting optimisation model are obtained via machine learning. This data-centric modeling can be used to get more accurate models or can be used in cases where there is no theoretical knowledge available to build the model manually.

Read more about the workpackages.

Valuation

The insights, gathered from the research project Optimization for and with Machine Learning (OPTIMAL) will be tested on a variety of applications where the consortium members are already involved, including classification problems in the medical sciences, decision problems related to the UN World Food Programme, and routing of shared, self-driving cars.

Contact information

Project coordinator: University of Amsterdam
Main applicant: prof. Dick den Hertog