He received a PhD in Mathematics from MIT, and has Bachelor's degrees in
Mathematics, Physics, and Psychology from Cornell University.
He has had appointments at Columbia
University as
Assistant and Associate Professor (Computer Science, Mathematics), at Tufts University as Assistant Professor,
and at MIT as a graduate instructor. He has served as departmental
director of graduate studies at Boston
University. He
has approximately 70 publications in mathematics, statistics, computer
science, neural network theory, and mathematical physics, including one
book. His current research and applications interests involve
learning theory, statistics and probability, neural networks, complexity
theory, optimization, and mathematical physics. His recent work in
learning theory has investigated complexities of designs for learning
machines and neural networks which improve significantly on those for
standard backpropagation architectures. He is on the editorial board
of Neural Networks, and has been on the organizing committee of the
World Congress on Neural Networks twice. He has had recent research
grants and contracts from the American Fulbright Commission, National
Science Foundation, and the U.S. Air Force. He has given
approximately 100 lectures in 15 countries. Among organizational
roles, he has been a coorganizer for MIT summer analysis seminars in Vermont, and the
organizer of a minisymposium on Computational Complexity Theory in Chamonix, France.

Research: Mark Kon works in machine learning, mathematical
neural network theory, complexity theory, statistical learning theory,
wavelets, and mathematical physics. His current research focuses on
learning as a statistical phenomenon in which an intelligent system learns
to combine a priori information with current data to form a model of an
inputoutput function to be learned. This area naturally connects to
complexity theory, neural network theory, and Bayesian inference, areas in
which similar issues are prominent. He and his coworkers focus on
connections between these approaches, and more generally on formulation of
an approach which unifies them. One major goal of this project is to
provide a normative index in which learning algorithms arising from various
approaches can be compared in a single setting.
Online Publications
Abstracts (in progress):
(in progress)
Number of
visitors since October
14, 2003
