Research at IVDA


Research at IVDA includes various methodologies and branches, all related to the meaningful combination of both strengths of humans and machines in the data analysis, knowledge generation, and decision-making process. Particular research tracks are as follows.


Interactive Data Labeling

The assignment of labels to data instances is a fundamental prerequisite for many machine learning tasks: only with labeled training data supervised machine learning is possible. We study the process of labeling data instances with the user in the loop, from both the machine learning (leveraging active learning) and the interactive and visual perspective (enabling humans to select instances by themselves). Our research builds upon our methodology “visual interactive labeling” (VIAL), which unifies both perspectives. VIAL describes the six major steps of the process and discusses their specific challenges.


Visual Analytics in Healthcare

The analysis of medical data is relevant for many use cases, including clinical research and patient treatment. While these cases typically represent a doctors'/physicians' perspective, there is yet another user group with an immense information need: patients. In addition, other types of stakeholders involved in the medical domain are, e.g., clinics, quality management environments, operational staff, and insurances.


Interactive Machine Learning

Interactive Machine Learning focuses on human-centered aspects of Machine Learning and the iterative machine learning process. Overall goal is to train powerful machine learning models by combining both the strengths of humans and machines. Interactive Machine Learning is in line with the Human-Centered Artificial Intelligence principle, fostering both a high human control and computer automation at the same time, to arrive at reliable, safe & trustworthy AI solutions.
class separation


Visual Analysis of Time Series

Time-oriented data (time series) are among the most prominent and frequently applied data types. Examples include EEG data in medicine, stock charts in finance, or temperature progressions in Earth observation. Time series come with special syntactic and semantic characteristics that require individual if no unique treatment. Special analysis tasks include the exploration of motifs/shapes, the identification of periodic/cyclic patterns, as well as the identification of trends. Also, visualization and interaction research has produced a series of unique designs for time series data.
time series search


Interactive Creation and Explanation of Item Rankings

Rankings of items enable users to a) select a top candidate item, b) focus on a small set of highly preferable items, or c) identify items that are ranked particularly weak.

The human-centered ranking of multidimensional items is a non-trivial task. Comparing complex items to each other in order to create a ranking can be time-consuming and, especially if data sets are large, there is no guarantee that the result is satisfying.

The scope of this project is two-fold. On the one hand, we characterize, design, and evaluate interactive solutions for the creation of item rankings. On the other hand, we design and develop ranking explainers, to enable users to gain more trust into given ranking models.

ASF teaser