Sandia Labs FY21 LDRD Annual Report

FY21 ANNUAL REPORT

Assessing cognitive impacts of errors from machine learning and deep learning models on human cognitive performance. Researchers characterized the impact of errors from machine learning and deep learning algorithms on human cognitive performance in visual search and identification tasks. The results support a systems- engineering approach to implementating these algorithms by quantifying the impact of different types, rates, and visual implementation (e.g., explainability or confidence metrics) of errors and providing evidence-based recommendations for integrating machine learning and deep learning with human decision making within analytical systems. The research focused on common visual tasks (search and identification in still images) and produced results that will generalize across multiple national security domains, including international nuclear safeguards, physical protection, intelligence, homeland security, and nuclear weapons manufacturing/engineering. (PI: Zoe Gastelum) Anticipating group dynamics through emergent recursive multiscale interaction. Understanding group emergence can improve our ability to anticipate and understand how group dynamics can be influenced in national security applications. Particularly impactful is the new theory on the multiscale, fractal-like nature of the emergence of groups, including recursive interactions between scales that resulted from this LDRD project. This work, supported through collaborations with professors Yuguo Chen at University of Illinois Urbana-Champaign and Abdullah Mueen at the University of New Mexico, will be important to NNSA and the DoD. (PI: Asmeret Naugle)

Group dynamics modeled using new communication vibration theory: (1) Movement of members between days; (2) changes in the average distance between group centroid and members between days; and (3) movement of centroid of group members relative to non-members.

51

LABORATORY DIRECTED RESEARCH & DEVELOPMENT