U.S. Department of Energy

Pacific Northwest National Laboratory

Transpire: Transparent Model-Driven Discovery of Streaming Patterns

In streaming data environments, the only thing that is constant is change. To the human observer or analyst, changes are often too fast to notice, too many to remember, and too complex to understand or make predictions. Automated methods such as data mining, machine learning, etc., can help in extracting and presenting meaningful information to the user for real-time decision-making. However, the outputs of such analytical models often suffer from lack of interpretability and need human feedback for achieving greater accuracy. To help analysts discover and reason, we will be conceptualizing a transparency-based visual analytics framework for integrating visualizations with automated methods that increase confidence and accuracy of human judgment for streaming data analysis.


Most existing visual analytics techniques are based on assumptions about how human-stream feedback should work and are not supported by concrete empirical evidence. In our work we fill these gaps through three main contributions:

  • Test how varying levels of model transparency can best leverage the high bandwidth of the human perception system for communicating key insights.
  • Study how differences in model transparency affect efficiency, effectiveness and the trustworthiness of human insights.
  • Aim to demonstrate the efficacy of our methods through case studies and quantitative user studies that demonstrate how these visual analytics methods can significantly improve human interpretability of machine detected streaming patterns.


A transparency-based visual analytics framework will benefit both experts trained in analytical methods whose goal is to build more accurate models about the data and domain experts who are consumers of the insights generated by these methods. The Transpire project will lead to a thorough characterization of the trade-offs for design and evaluation of streaming visual analytics techniques, thereby advancing the science of human reasoning about data streams.

| Pacific Northwest National Laboratory