Visual-Interactive Machine Learning for Time-Oriented Data
Event date:  January 29, 2018 4:00 PM  to 5:00 PM

Talk Abstract

In the last years, the Visual Analytics (VA) community has made a considerable step towards Machine Learning (ML). Conceptual models for knowledge generation have been proposed that seek to combine VA and ML, VA approaches have been presented that make massive use of ML techniques in a visual-interactive way. Finally, real-world applications (involving domain experts from various application areas) have been designed, making use of the combined strengths of VA and ML.

In my talk, I will share my experiences on the conflation of ML with VA, by the example of time-oriented data. Time-oriented data is a data type with the characteristics that every value depends on a specific point in time. This characteristics allows to store special natural or human-made phenomena. As such, time-oriented data has special analytical potential for researchers, engineers, or doctors (to name experts groups of my personal collaboration record), but also has special challenges that we are confronted with. After several collaborative projects, a PhD thesis, and a considerable amount of publications that involved time-oriented data, I will reflect on visual-interactive machine learning for time-oriented data in two parts.

First, I will talk about exploratory search, a concept that combines two complementary analysis tasks with the goal to create increased analytical benefit in the knowledge generation process. I will emphasize three aspects that allow the combination of VA and ML: (i) preprocessing, (ii) clustering and dimensionality reduction, and (iii) seeking relations between data content and metadata.

Second, I want to share research results for the segmentation and labeling of time-oriented data. Segmentation is the principle to divide complex (time-oriented) data into yet more meaningful units, while labeling refers to the principle to create meaning to data, e.g., to segments. Again, combinations of VA and ML techniques have proven to be beneficial to achieve useful solutions.

Speaker’s Bio

Jürgen Bernard is Post-doc at the department of Computer Science, TU Darmstadt, Germany. He is leading the Visual-Interactive Machine Learning research group in the Interactive Graphics Systems Group (GRIS). He was with Fraunhofer IGD when he received is PhD degree in 2015, entitled “ Exploratory Search in Time-Oriented Primary Data”.

In 2016, his work was awarded with the Hugo-Geiger Prize for excellent PhD theses. His research is in information visualization, visual analytics, machine learning, and user-centered design and addresses a broad range of application fields with an emphasis on medical health care and data-driven human-centered health. In 2017, J. Bernard received the Dirk Bartz Prize for Visual Computing in Medicine from the Eurographics Association (EG).

His data-centered focus is on multivariate data, multimodal data, and time-oriented data. Many of his visual-interactive solutions combine techniques from cluster analysis, dimensionality reduction, similarity search, active learning, or other approaches related to information retrieval, data mining, and machine learning.

His publication list contains over 60 entries, more than half authored by J. Bernard in the leading position. His publication list includes more than ten journal publications, over 30 conference publications (and conference participations), overall with almost ten award-bringing entries.



University of Konstanz, Universitätsstr. 10, Konstanz
Powerwall C202 

University of Stuttgart, VISUS-Building, Allmandring 19, Vaihingen
Powerwall Room -01.116 (Live Transmission)