In the NWO Open Competition ENW-GROOT programme, four CWI researchers received in total two grants to study machine learning and neural networks. Nikhil Bansal, Monique Laurent and Leen Stougie are involved in the project ‘Optimization for and with Machine Learning’, and Benjamin Sanderse in ‘Unravelling Neural Networks with Structure-Preserving Computing’. The NWO Open Competition ENW-GROOT programme is intended for consortia in which research groups create added value through collaboration.
Optimization for and with Machine Learning
Machine learning is often in the news because of remarkable applications such as image recognition and self-driving cars. When constructing machine learning models, such as deep learning and random forests, mathematical optimization plays an important role. In the first project part we want to better understand the performance of existing optimization techniques for machine learning and also develop faster and better optimization techniques. In the second part we use machine learning techniques to solve optimization problems faster and more accurately. The new techniques are applied to classification problems for medical treatments, finding genetic relationships, food distribution chains for the World Food Programme, and self-driving cars.
This programme is coordinated by Tilburg University, with participants at TUD, Tilburg University and CWI. Nikhil Bansal and Monique Laurent from CWI's Networks and Optimization group say: "At CWI, we will study combinatorial and polynomial optimization based methods for machine learning, develop theoretical analysis for heuristics used in machine learning, and use machine learning to design optimization algorithms that exploit structure in data". Leen Stougie from CWI's Life Science and Health group adds: "In close collaboration with TUD we will work on improving models and optimization methods for life sciences inspired problems, like classification of virulent yeast strains and problems in phylogeny".
Unravelling Neural Networks with Structure-Preserving Computing
Machine learning with neural networks revolutionizes our daily lives such as the automation of complex tasks such as speech recognition. It also finds its way into the simulation of phenomena in physics, chemistry, astronomy and biology. For the latter, a better understanding is essential in order to construct efficient, tailor-made neural networks using properties of the underlying scientific problems. The resulting deeper understanding of neural networks from a mathematical, physical and astronomical perspective is vital for future developments in this rapidly evolving field.
This programme is coordinated by Eindhoven University of Technology, with participants at Leiden University, TU/e and CWI. Benjamin Sanderse from CWI's Scientific Computing group explains: “We will investigate how constraints can be included into neural network architectures by interpreting neural networks as systems of constraint differential equations.”
In the NWO Open Competition ENW - GROOT programme of this year, twenty new consortia are starting a major research project. The impulse of more than 47 million euro enables new curiosity-driven, independent fundamental research.
More information
Optimization for and with Machine Learning (OPTIMAL)
- D. den Hertog, Tilburg University (coordinator)
- L. Stougie (CWI)
- E. de Klerk (Tilburg University)
- M. Laurent (CWI)
- K.I. Aardal (TUD)
- N. Bansal (CWI)
- L.J.J. van Iersel (TUD)
Unravelling Neural Networks with Structure-Preserving Computing
- W.H.A. Schilders (TU/e) (coordinator)
- B. Sanderse (CWI)
- S.F. Portegies Zwart (UL)
- B. Koren (TU/e)
- F. Toschi (TU/e)
- J.W. Portegies (TU/e)
