Jay McClelland
Thursday 25th June 2015
Time: 4.00pm
Basement Seminar Room
Alexandra House, 17 Queen Square, London, WC1N 3AR
A Parallel Distributed Processing Approach to Mathematical Cognition
The Parallel Distributed Processing Approach to modeling cognition
challenges the view that the brain uses explicit symbolic
representations to capture the knowledge underlying our natural
cognitive abilities, including our ability to perceive, understand, and
produce language, our intuitions about the properties of objects, and
even our intuitions about physical principles, such as the role of
weight and distance from a fulcrum in determining which side of a
balance scale will go down. According to the PDP approach, the knowledge
underlying these natural cognitive abilities is stored in the strengths
of connections among simple processing units, and learning occurs
through a process of connection adjustment, driven by experience.
Knowledge stored in this way is not accessible as such for overt report,
but can govern quite complex intuitions in all of the indicated domains,
as cognitive modeling and machine learning research makes clear. But
can this approach tell us anything about how people solve problems in
mathematics, or how children acquire mathematical abilities? In a new
research direction, I have begun to explore this topic. This new
direction is beginning to bear fruit, as I will explain by describing
new research projects from my laboratory.
James L. McClelland, Department of Psychology and Center for Mind,
Brain, and Computation, Stanford University