Titel: The role of concentration inequalities for learning from (non-) i.i.d. data
Abstract: The analysis of modern machine learning algorithms require a variety of techniques from different mathematical areas. As an introduction to this field, I will therefore first provide an overview over notions, questions, and methods considered in statistical learning theory. The main part of my talk will then focus on the statistical part of the analysis. Here, I will first illustrate the main effects, different types of concentration inequalities have on the statistical analysis if the data is i.i.d. I will then show, how concentration inequalities for non-i.i.d. data can be employed in the analysis, and which difficulties need to be addressed. In the final part, I will present some examples, in which the step from i.i.d. to non-i.i.d. was either successful or not.