Title: Interactive Visualization to Externalize, Explore, and Explain Trust in ML

Abstract: The increased availability of large data sets and novel computational analysis methods greatly increases the complexity of experts' data analysis and decision-making processes. Our approach to dealing with these challenges comes from the field of Visual Analytics, defined as "the science of analytical reasoning facilitated by interactive visual interfaces". Visual Analytics sets out a broadly interdisciplinary human-centered design process to create interactive graphical information environments to improve analysis and to demonstrate the rigour of analysis and decision-making processes  (e.g. "due diligence") for community engagement, stakeholder consultation, and regulatory or legal review.

The addition of machine learning methods for data analysis creates a new challenge, requiring us to find new ways to extend VA methods to include ML processes and outcomes. In 2022 members of the VA community organized a Dagstuhl Seminar on "Interactive Visualization for Fostering Trust in ML". Our group focused on development of a conceptual framework for extending visual analytics to formalize and externalize trust in machine learning processes. My talk will describe this approach and discuss how it might be helpful in a range of application projects.