An Incremental Machine Learning Algorithm for Nuclear Forensics

Abstract of the technical paper presented at:
Canadian Conference on Artificial Intelligence
May 8–11, 2018

Prepared by:
Chris Drummond
National Research Council
Acknowledgement:
Domain knowledge and leadership through Ali El-Jaby (CNSC)

Background:

This paper presents work undertaken as part of the Nuclear Material Signature and Provenance Assessment Capability Development Project (NMS/PAC). The NMS/PAC is a whole-of-government Research & Development initiative led by the CNSC aimed at developing, enhancing and expanding Canada’s radioactive and nuclear material metrology and data analytics capabilities to support provenance assessment functions for nuclear forensics operations. NMS/PAC partners include: National Research Council, Atomic Energy of Canada Ltd. /Canadian Nuclear Laboratories and the University of Ottawa.

The NMS/PAC is supported in part by the Canadian Safety and Security Program (CSSP), which is led by Defence Research and Development Canada’s Centre for Security Science, in partnership with Public Safety Canada.

The CSSP is a federally-funded program to strengthen Canada’s ability to anticipate, prevent/mitigate, prepare for, respond to, and recover from natural disasters, serious accidents, crime and terrorism through the convergence of science and technology with policy, operations and intelligence.

Abstract:

This paper presents an incremental machine learning algorithm that identifies the origin, or provenance, of samples of nuclear material. This is part of work being undertaken by the Canadian National Nuclear Forensics Library development program, which seeks to build a comprehensive database of signatures for radioactive and nuclear materials under Canadian regulatory control. One difficulty with this application is the small ratio of the number of examples over the number of classes. So, we introduce variants to a basic generative algorithm, based on ideas from the robust statistics literature and elsewhere, to address this issue and to improve robustness to attribute noise. We show experimentally the effectiveness of the approach, and the problems that arise, when adding new examples and classes.

To obtain a copy of the abstract’s document: https://nparc.nrc-cnrc.gc.ca/eng/view/object/?id=b6778d85-3f82-44e4-9e01-8ef0c950a9a0