Oct 23 2020
Ensuring data quality is critical for artificial intelligence (AI) machines to learn effectively and operate efficiently and safely.
Researchers from Florida Atlantic University's College of Engineering and Computer Science have received a three-year, $653,393 grant from the United States Air Force Office of Scientific Research (AFOSR) for a project titled, "Data Analytics and Data Conformity Evaluation with L1-norm Principal Components." For the project, researchers will develop new theory and methods to curate training data sets for AI learning and screen real-time operational data for AI field deployment.
The project team is spearheaded by Dimitris A. Pados, Ph.D., principal investigator, a professor in the Department of Computer and Electrical Engineering and Computer Science, a fellow of FAU's Institute for Sensing and Embedded Network Systems Engineering (I-SENSE), the Charles E. Schmidt Eminent Scholar in Engineering and Computer Science, and director of the Center for Connected Autonomy and Artificial Intelligence (ca-ai.fau.edu) who is nationally renowned in the areas of machine learning and connected AI.
This latest project for AFOSR will involve basic research to develop novel mathematical methods that measure the conformity of individual data points with respect to all others available, in a blind, unsupervised manner. The developed mathematical data-conformity evaluation schemes will process any given data set represented by a high-dimensional matrix (also known as tensor) and convert each data entry to a continuous zero-to-one "alert conformity value" (zero implying highly-conforming data; one implying highly non-conforming data).
"AI systems learn from examples and the quality - correctness and completeness - of the set of examples, presented to an AI machine to learn from, and this is obviously a core matter in AI technology. After training, the AI machine will eventually be let out 'in the wild' to operate on its own - autonomously - on fresh real-time sensed data," said Pados. "Non-conforming sensed data may represent critical and actionable information like an internal system or sensor failure. Our ability to identify this 'non-conforming' data could potentially make AI and autonomously operated systems safer than human-operated machines of present time."
AFOSR accomplishes its mission by investing in basic research efforts for the Air Force in relevant scientific areas, which include engineering and complex systems; information and networks; physical sciences; and chemistry and biological sciences.
"With this important grant from the United States Air Force Office of Scientific Research, professors Pados and George Sklivanitis and the project team will develop cutting-edge technology to identify faulty, unusual and irregular information for AI learning and operations that rely on data, and will provide critical alerts to troubleshoot a problem before it occurs," said Stella Batalama, Ph.D., dean of FAU's College of Engineering and Computer Science. "This data-quality evaluation technology is being developed for applications in a number of industries ranging from the military to cybersecurity to medical diagnostics."