Machine Learning (iCIS)

The high-level ambition of the Machine Learning group at the Institute for Computing and Information Sciences is to work towards "bounded rational" machine learning: machine learning methods that properly take into account all information available (both data and prior knowledge) and then provide an optimal answer, given finite resources and computation time.

The Machine Learning group aims to show that such algorithms (indeed) improve the state-of-the-art, both in theory and in practice. By applying machine learning methods to problems in other scientific domains, such as (cognitive) neuroscience and bioinformatics, the machine learning group wishes to contribute to the progress in these other disciplines.

We follow two approaches: top-down, through a Bayesian approach, and bottom-up, by designing clever heuristics. The Bayesian approach has - in theory - many desirable properties, such as consistency, coherence, and optimality, but in practice leads to computationally intractable problems. More heuristic approaches are designed to be tractable in the first place, but are more difficult to interpret in terms of objective measures of optimality. Having expertise on both sides, we strive to combine the best of both worlds.

We hook up with other scientists, at the Donders Institute and elsewhere, to work together on the analysis of their data to extract new scientific knowledge. This does not only provide a test-bed for existing machine learning methods, but also provides inspiration for new ones.

Links

• Website Machine Learning Group