EuroVA19
Permanent URI for this collection
Browse
Browsing EuroVA19 by Author "b7736374-f807-4f21-82f0-e4a978fec7d3"
Now showing 1 - 2 of 2
Results Per Page
Sort Options
Item Quantifying Uncertainty in Multivariate Time Series Pre-Processing(The Eurographics Association, 2019) Bors, Christian; Bernard, Jürgen; Bögl, Markus; Gschwandtner, Theresia; Kohlhammer, Jörn; Miksch, Silvia; Landesberger, Tatiana von and Turkay, CagatayIn multivariate time series analysis, pre-processing is integral for enabling analysis, but inevitably introduces uncertainty into the data. Enabling the assessment of the uncertainty and allowing uncertainty-aware analysis, the uncertainty needs to be quantified initially. We address this challenge by formalizing the quantification of uncertainty for multivariate time series preprocessing. To tackle the large design space, we elaborate key considerations for quantifying and aggregating uncertainty. We provide an example how the quantified uncertainty is used in a multivariate time series pre-processing application to assess the effectiveness of pre-processing steps and adjust the pipeline to minimize the introduction of uncertainty.Item Visual Analysis of Degree-of-Interest Functions to Support Selection Strategies for Instance Labeling(The Eurographics Association, 2019) Bernard, Jürgen; Hutter, Marco; Ritter, Christian; Lehmann, Markus; Sedlmair, Michael; Zeppelzauer, Matthias; Landesberger, Tatiana von and Turkay, CagatayManually labeling data sets is a time-consuming and expensive task that can be accelerated by interactive machine learning and visual analytics approaches. At the core of these approaches are strategies for the selection of candidate instances to label. We introduce degree-of-interest (DOI) functions as atomic building blocks to formalize candidate selection strategies. We introduce a taxonomy of DOI functions and an approach for the visual analysis of DOI functions, which provide novel complementary views on labeling strategies and DOIs, support their in-depth analysis and facilitate their interpretation. Our method shall support the generation of novel and better explanation of existing labeling strategies in future.