Abstract
The notion of uncertainty is of major importance in machine learning andconstitutes a key element of machine learning methodology. In line with thestatistical tradition, uncertainty has long been perceived as almost synonymouswith standard probability and probabilistic predictions. Yet, due to thesteadily increasing relevance of machine learning for practical applicationsand related issues such as safety requirements, new problems and challengeshave recently been identified by machine learning scholars, and these problemsmay call for new methodological developments. In particular, this includes theimportance of distinguishing between (at least) two different types ofuncertainty, often refereed to as aleatoric and epistemic. In this paper, weprovide an introduction to the topic of uncertainty in machine learning as wellas an overview of hitherto attempts at handling uncertainty in general andformalizing this distinction in particular.