Abstract
Traditional data mining and machine learning assume that data comes from a fixed distribution, therefore, predictive models built on data are also fixed. But as the world is continuously changing, so does most of the data that describes it. In order to realistically capture real-world phenomena, data-driven models need to be able to adapt to evolving data, otherwise they may lose accuracy and become obsolete over time. Research attention to such modeling scenarios has been increasing in the last decade, a number of adaptive learning algorithms have been developed. This talk will overview research in adaptive learning from evolving data, and discuss an example study on methods for balancing resources in such learning.
Indrė Žliobaitė
University of Helsinki and Helsinki Institute for Information Technology HIIT
Bio
Indre Zliobaite is a researcher at University of Helsinki, and Helsinki Institute for Information Technology HIIT. Her main research interests include learning from evolving data, adaptation, transparency and accountability in machine learning, and computational data analysis in general.