Uncertainty Quantification in Machine Learning - From Aleatoric to Epistemic

Explore the critical concepts of uncertainty quantification in machine learning through this comprehensive lecture delivered by Professor Eyke Hüllermeier from LMU Munich at the RWTH Center for Artificial Intelligence. Delve into the fundamental distinction between aleatoric uncertainty, which stems from inherent randomness in data generating processes, and epistemic uncertainty, which arises from the learner's incomplete knowledge of the true underlying model. Learn about the representation and appropriate handling of predictive uncertainty in supervised machine learning contexts, with particular emphasis on developing numerical measures to quantify these different types of uncertainty. Examine the conceptual and theoretical challenges associated with existing uncertainty quantification methods, understanding why disentangling aleatoric and epistemic uncertainty remains a complex problem in machine learning. Gain insights from Professor Hüllermeier's extensive research experience in artificial intelligence, machine learning, and reasoning under uncertainty, as he addresses the growing importance of uncertainty quantification driven by safety requirements in practical applications. Discover the theoretical foundations and methodological approaches that are essential for developing reliable machine learning systems capable of properly handling and communicating uncertainty in their predictions.