KIU-2021 Keynote: Uncertainty in Machine Learning for Environmental Research

Ludwig, Nicole2021-12-142021-12-142021978-3-88579-708-1https://dl.gi.de/handle/20.500.12116/37681Advances in metering infrastructure, data collection and storing have led to data-driven research approaches such as machine learning to become more revalent in environmental and sustainability research. However, probabilistic approaches which handle uncertainty in the available data and models are still underrepresented. This underrepresentation is prevalent in environmental research, in addition there are different uncertainties influence the models we design and their output. For example, the data sets we have at hand are not always equally well distributed in space and time; some areas might have lots of data points (high resolution in space) while others might have more frequent data (high resolution in time). When working with a mixed data set, probabilistic measures can help clarify where the models are certain about the results and where they are uncertain. Additionally, many research questions we want to answer, such as the optimal location for a wind turbine, rely on weather and climate data input. As weather and climate are a chaotic system and thus only predictable to a certain extent, accounting for this uncertainty is crucial for reliable forecasts and any decision making based on them. In this talk, we will address some of the open issues with uncertainty in space and time in the input data and output of probabilistic machine learning models at the example of forecasting in sustainable energy systems. Frau Dr. Ludwig leitet die Early Career Forschungsgruppe für Maschinelles Lernen in Nachhaltigen Energiesystemen am Exzellenzcluster Maschinelles Lernen der Universität Tübingen. Ihre Forschungsarbeiten umfassen die Entwicklung von Algorithmen des Maschinellen Lernens für Probleme im Zusammenhang mit zukünftigen, nachhaltigen Energiesystemen mit hohem Anteil erneuerbarer Energiequellen. Ihr besonderes Interesse gilt dem Probabilistischen Lernen auf Zeitreihen und dem Bestärkenden Lernen mit unsicheren Daten. Frau Ludwig hat am KIT in Informatik promoviert und besitzt einen MSc in Information Systems and Network Economics sowie einen BSc in Economics der Universität Freiburg.enKIU-2021 Keynote: Uncertainty in Machine Learning for Environmental Research10.18420/informatik2021-0191617-5468