Modern science, technology, and politics are all permeated by data that comes from people, measurements, or computational processes.  However, data is often incomplete, corrupt, or lacking in sufficient accuracy and precision.  While concern for these uncertainties would seem essential to rational decision making, explicit consideration of uncertainty is rarely part of the computational and decision making pipeline.

To address this critical shortcoming in the way we process, present, and interpret data, significant improvements are needed both in the modeling of computational uncertainty and in the tools used to communicate uncertainty to decision makers.  Success will require a broad based multidisciplinary effort, involving development of a comprehensive set of foundations for representing and communicating computational uncertainty that accounts for all aspects of the problem—including the applications, the numerics, the visualizations, and the comprehension of users—in a holistic, systematic manner.

Workshop Synopsis:

Now is the appropriate time to hold a discussion about future research directions related to the modeling of uncertainty in computations and the ways in which the uncertainty inherent in many computational processes can be communicated to those tasked with making decisions based on such data. Uncertainty quantification for computational simulations is a maturing discipline, but little study has yet gone in to the relationship between uncertainty quantification and the communication of uncertainty to decision makers. Data analytics is rapidly becoming far more sophisticated and enjoying widespread use, but is still largely lacking in well principled methods for quantifying uncertainty associated with the information contained in large data sets.  The field of decision science recognizes the importance of understanding decision making under uncertainty, but much of this work is not closely integrated with either formal uncertainty quantification or the explosion of computational uncertainty associated with data analytics.

The workshop addressed these issues in two ways.  A set of research challenges will be defined that, if solved, will make the computation and utilization of uncertainty more ubiquitous in a variety of computing applications and systems.  In addition, joint goals and methods between different disciplines will be identified to help establish an interdisciplinary agenda for addressing challenges that uncertainty poses.  Success in these efforts will accomplish better decision making through a better understanding of uncertainty, better understanding of models and their accuracy by data analysis and simulation scientists, and increased credibility of computational estimates and simulations by the public through better understanding of uncertainty.

Workshop Outcomes:

The most immediate outcome of the workshop was a set of white papers describing the specific aspects of the research agenda arising out of the discussions at the workshop, along with a workshop report.

Over the longer term, workshop organizers will be engaging funding agencies about the importance of the problem of computational uncertainty and the proposed research agenda.