Human-Centered Machine Learning is about closely linking the possibilities of human perception and intelligence with the computing capacity and performance of computers. However, there are two basic views of integrating humans in the ML process.
On the one hand, the human being is understood as a component of the technical system (human-in-the-loop), which can optimize the learning behavior of algorithms through its interactions with the system. Approaches in this area include reinforcement learning, preference learning, and active learning.
However, in this seminar, we will focus on another perspective. Human-Centered Machine Learning is based on the idea that ML should also be usable by non-technical experts, i.e., persons without background knowledge or experience in this field (algorithm-in-the-loop). User interface design is fundamental to the success of this perspective, but there is a lack of consolidated principles on how such interfaces should be designed. The transparency of such applications, for example, and interpretability of results are an essential prerequisite.
In this seminar, we will conduct a detailed review of existing approaches to human-centered machine learning from the perspective of interactive systems to contextualize them in the field of human-computer interaction. Based on an introduction of the various aspects of HCML, we learn about different approaches of integrating user interfaces in the ML process. Building on this, we open up this new field of research by a more reflective perspective on existing methods through student presentations.
Participants of this seminar are expected to prepare and present one given topic and to discuss the insights with the group. Based on the results of the discussion, participants will elaborate (paper) on their topic in more detail.