Sherpa: Leveraging User Attention for Computational Steering in Visual Analytics

We present Sherpa, a computational steering mechanism for progressive visual analytics that automatically prioritizes computations based on the analyst's navigational behavior in the data. The intuition is that navigation in data space is an indication of the analyst's interest in the data. Sherpa implementation provides computational modules, such as statistics of biological inferences about gene regulation. The position of the navigation window on the genomic sequence over time is used to prioritize computations. In a study with genomic and visualization analysts, we found that Sherpa provided comparable accuracy to the offline condition, where computations were completed prior to analysis, with shorter completion times. We also provide a second example on stock market analysis.