Machine Learning with Interdependent and Non-identically Distributed Data (Dagstuhl Seminar 15152)

Abstract

One of the most common assumptions in many machine learning and data analysis tasks is that the given data points are realizations of independent and identically distributed (IID) random variables. However, this assumption is often violated, e.g., when training and test data come from different distributions (dataset bias or domain shift) or the data points are highly interdependent (e.g., when the data exhibits temporal or spatial correlations). Both scenarios are typical situations in visual recognition and computational biology. For instance, computer vision and image analysis models can be learned from object-centric internet resources, but are often rather applied to real-world scenes. In computational biology and personalized medicine, training data may be recorded at a particular hospital, but the model is applied to make predictions on data from different hospitals, where patients exhibit a different population structure. In the seminar report, we discuss, present, and explore new machine learning methods that can deal with non-i.i.d. data as well as new application scenarios.