A sample of ongoing and published research spanning knowledge graphs for analysts, explainable AI, and human-in-the-loop machine learning. For the complete record, see Publications.
Establishing the effects of providing human-in-the-loop feedback on user perceptions and behavior in differing contexts, showing negative effects on trust and system perceptions when providing objective feedback and positive effects when providing reasoning-based feedback.
A DARPA XAI study on how explanations from a dynamic cutset network affect trust in video activity recognition. Evaluation shows that early impressions of an AI system's mistakes anchor users' trust well after true accuracy becomes clear.
Automated segmentation and summarization of analytic provenance data for handoff between analysts.
Explaining how human feedback is being used in a knowledge graph recommendation system.