Home
Research
Thirty years ago Eleanor Gibson
said that “learning is the education of attention.” How does attention
get educated?
Learning is about reducing error, or, in general, uncertainty. We
need to learn because we're not equiped out-of-the-shell to handle
every contingency the environment might deliver. Our human
ability to alter and construct environments, and our strong reliance on
social information, makes us very flexible learners.
Flexibility without constraint is chaos, and while chaos might be
fascinating it doesn't make for very reliable conversation.
To build a
reliable system in a flexible environment is an enviable ability,
understanding it would be a triumph. My goal is to travel along
this path; building, testing and rejecting models of human learning.
As Mackintosh noted in 1997, the last fifty years has seen a slow circle closing, in which modern learning theory, much of it associative, has returned to the fore in psychology. Connectionism is the prime example of this movement, and much of my research has used connectionist models. Connectionist networks learn by adjusting their weights; weights on input units and weights on hidden units. These weights effectively represent the attention the network pays to its inputs, or to complex properties of its inputs. What a connectionist network learns is to educate its attention.
``Attention'' is a word with many meanings, and to say that learning is an adaptive change in attention is necessarily vague. Indeed, this vagary is reflected in the claim that connectionist networks can model cognition; many have observed that connectionist systems are super-Turing machines, and so not sufficiently constrained to count as theories. The focus of my research continues to be to theoretically identify specific meanings of the word ``attention'', quantitatively model these meanings as mathematical systems, and empirically validate these models in psychological experiments.