Motivation and Goals

Printer friendly version of the Call For Papers

In the Data Science era, analyzing large amounts of abstract data through their graphical representation is a daily business for data scientists in various domains like Computational Biology, Bioinformatics, Social Computing, Cyber-Security, Finance, or High-Performance Computing. When visually analyzing their data, analysts are often overwhelmed with the sheer number of possible node-link diagrams, scatterplots or treemaps to look at, while not having a clear notion of which one might reveal interesting patterns. In this workshop, we will focus on how Computer Vision and Pattern Recognition can help to automatically filter out irrelevant views and put forward most promising ones. While this question is gaining more and more attention in the Visual Analytics community, we envision a large benefit by more actively including the CVPR community into these research endeavors. The goal of the workshop is therefore to bridge the gap between these communities and inspire more joint research projects.


Visual Analytics is a quickly growing field which aims at supporting data analysts with interactive and graphical representations of the data designed to increase insight discovery and complement traditional Machine Learning model-based analysis. During the Visual Analytics process, the analyst has to identify relevant representations of the data among a variety of possible graphical depictions. The space of possibilities includes the mappings on graphical variables used to encode the information (e.g. the form of the representation as scatterplots, node-link diagrams, parallel coordinates, treemaps... rendered as an image on the screen), as well as the features and samples, or their summary which are to be visualized (the underlying content of the representation).

Which representation leads to meaningful and interesting insights strongly depends on the abstract patterns contained in the data (correlation, clusters, outliers...), the ability of the representation to make these patterns salient to the human visual system, as well as user’s specific goals and prior knowledge.

In practice, analysts rely on their experience to select interesting images within this search space. They can spend most of their time exploring a large number of uninteresting views, testing different parameter settings by hand, and selecting or summarizing features and samples, until interesting results are visually identified. The identified insights enrich the analyst’s knowledge of the data, and further support the identification of other views of interest, based on the patterns previously seen.

Topics of Interest

The CV4VA Workshop invites researchers in Computer Vision, Pattern Recognition and Machine Learning to discuss and propose innovative ways to use advanced computing techniques for guiding the Visual Analytics process.

The issues we wish to handle include, but are not limited to:

  • What are the low and high level visual patterns or features used by the human visual perception system when visualizing graphical representations like scatterplot, parallel coordinates plots, node-link diagrams, treemaps, or other visualizations?
  • How can these visual patterns or features be leveraged to design new computer vision algorithms for automatic extraction, similarity measurements and analysis of patterns in such graphical representations?
  • How can similarity measurements or specific pattern discovery tasks, such as outlier detection, class separation, or cluster detection, be efficiently automatized in given standard graphical representations?
  • How to exploit these models to recommend relevant graphical representations to data analysts?
  • What are meaningful human-labeled visualizations (i.e., images with human judgment or tagging) in the context of Visual Analytics and how to collect large collections of them?
  • How can we use these labeled images to model human perception and possibly higher level cognition, improve computer vision and pattern recognition techniques in that context and finally quantify and improve the visual analysis experience?