The colloquia (unless otherwise stated) will take place on Tuesdays at 3.00pm in Room 500 in the Mathematics Department. See the map for further details. There will be a small reception afterwards in Mathematics Room 502 (25 Gordon Street - see the map). If you require any more information on the Departmental Colloquia please contact Prof Dima Vassiliev e-mail: d.vassiliev AT ucl.ac.uk or tel: 020-7679-2442.
23 April 2019
Speaker: Jürgen Jost (Director, Max Planck Institute for Mathematics in the Sciences, Leipzig)
Title: How to find structure in data? Geometry, heuristics, and data analysis
When confronted with large, quickly changing data sets from perhaps unknown sources and of varying reliability and quality, both humans and machine learning algorithms need to make certain structural assumptions in order to make sense of them. These assumptions are usually of a geometric nature, like smoothness, symmetry, hierarchical structure or few sources only.
These often translate into powerful heuristics for humans and efficient machine learning algorithms for computers. A deeper understanding is a mathematical challenge at the interface of high dimensional geometry, discrete mathematics, statistics, information theory and other fields. I shall discuss some examples and try to outline some perspectives.
Jürgen’s talk will be followed by a presentation led by David Tuckett (UCL) from the CRUISSE network (Challenging Radical Uncertainty in Science, Society and the Environment)
The decision-making challenges facing Government and Business are complex. We use the term radical uncertainty to refer to contexts in which, at the time you are making decisions, you cannot know if what you are deciding will work out as expected. Many real-world decisions are of this type and David Tuckett (UCL Sociology and Psychology), Lenny Smith (LSE, Maths and Physics), David Good (Cambridge, Psychology and Design) and Timo Ehrig (MPI MiS, Maths) will briefly explore the implications as to when and how far it is safe to use the modelling techniques that, to date, have formed the backbone of decision science.