Mathematics Influences Neural Research on “Deep Learning”

Posted Sep 29, 2020

Deep neural networks are likened to vast, complex computing systems and can be analyzed using formal methods and techniques. In recent years, Artificial Intelligence (AI) has changed its focus from fundamental decision-making to rigorous mathematical reasoning. As a result, the Department of Defense and The Office of Naval Research has granted a team of mathematicians, engineers and statisticians more than $7 million over the next five years to study artificial neurons as they navigate and process information to make improper decisions.

Although this may seem contrary to what the researcher should be studying, the nature of mathematical proof is to purposefully find what doesn’t work. We use null hypotheses to test against what we think may be probable. Mathematical science relies heavily in non-examples in order to create concepts that are well-defined. For these reasons, the research consortium from Johns Hopkins, Texas A&M, University of Maryland, the University of Wisconsin, UCLA and Carnegie Mellon are attacking the problem of untangling “deep learning” from three distinct perspectives: mathematics, statistics and formal verification. 

To read more about their work and how mathematics is directly influencing contemporary AI models, visit here.

MAEM Graduate Director Jeff Smith