TITLE: Scenario Reduction and Clustering Revisited: Fundamental Limits and Gurarantees

ABSTRACT:

The proliferation of “big data" created by mobile devices, social media, sensor networks, satellites etc. offers the potential to make more informed decisions, but it also poses a significant challenge to the scalability of the current optimization methods and decision support tools. This raises a number of fundamental questions: How should uncertain parameters be represented so that they lend themselves to integration into optimization problems? Which observations and features of a high-dimensional parameter set are most relevant for a particular optimization problem? How can these relevant observations and features be identified efficiently? How can one maintain the computational tractability of the emerging optimization problems? In the first part of the talk I will introduce tractable approaches for reducing large datasets to smaller ones at minimal loss of information. In the second part I will describe tractable methods for recognizing balanced clusters of similar datapoints in large unstructured datasets. Both methods come with rigorous performance guarantees. Implications for practical decision-making will be highlighted.

BIO: Daniel Kuhn holds the Chair of Risk Analytics and Optimization at EPFL. Before joining EPFL, he was a faculty member at Imperial College London (2007-2013) and a postdoctoral researcher at Stanford University (2005-2006). He received a PhD in Economics from the University of St. Gallen in 2004 and an MSc in Theoretical Physics from ETH Zurich in 1999. His research interests revolve around robust optimization and stochastic programming.