EAGER: Time Aware Research Community Identification of Critical & Emerging Technologies: Global Multi-sector Quantum Information Science (QIS)

Innovation in science and engineering contributes significantly to the nation’s economic growth, global competitiveness, and national security. The federal government identifies key areas of research, “Critical and Emerging Technologies (C&ET),” that are of particular importance to the nation. C&ET fields like Artificial Intelligence, Nuclear Energy, Quantum Computing, and Biotechnology are also areas of international competition. In recent years the federal government has taken steps to improve international research security to protect C&ET fields. Because C&ET research tends to be interdisciplinary, fast moving and to take place in many types of organizations – universities, companies, research institutes, national labs - it can be very difficult to identify all the research and researchers working in a particular field. This pilot project uses artificial intelligence (AI) and computational network methods and huge (100+ million) datasets of international patents, publications and grants to test new approaches to identifying critical and emerging technology researchers and research in Quantum Information Science (QIS) as a first step toward developing a validated approach that can be used for any critical and emerging technology to support research and evaluation of research security.

This Early Concept Grant for Exploratory Research (EAGER) award approaches the problem of C&ET identification by treating it as a problem of community identification in complex evolving networks. This project will develop time aware multidimensional representations of an evolving international network of researchers pursuing work related to QIS based on international patents, scientific publications and research grants drawn from Digital Science’s Dimensions database. A precision-maximizing “seed set” of known QIS researchers will include authors on papers published in specialized QIS journals and inventors on patents in relevant Cooperative Patent Classification codes issued to companies participating in the US National Quantum Initiative. This group of researchers will serve as a labelled subset for the development and testing of semi-supervised models to identify additional QIS researchers in a multi-dimensional network comprised of Large Language Model (LLM) vector embeddings of patent, paper and publication abstracts, graph embeddings of co-authorship and co-inventorship, citation and journal/CPC co-publication/invention networks. Exploratory model development will progress from simple label propagation to deep learning graph neural networks with convolutional and attentional layers. Inclusion of variational autoencoder layers will test the possibility that adversaries may obscure key network ‘signals’ in pursuit of malign goals. Ablation studies across network and model layers relative to multiple validation datasets will identify the most effective mix of model accuracy and computational cost. All aspects of model and data development will be configured to allow expansion to other C&ET areas creating the potential for generalization and use across the full range of fields in future research on research security.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Award Number: 2537124
Principal Investigator: Jason Owen-Smith
Funds Obligated: $299,840
State: MI