Department of Computer Science
Wednesday 4th August 1999 (6 pm)
ABSTRACT
The minds of adults and infants are so different that it is hard to imagine how one could develop from the other. Infants are sensorimotor agents that lack conceptual systems, symbolic representations, and the ability to reason. How do these abilities develop? We ask the same question of mobile robots, which have rudimentary sensors, effectors and reflex-like actions, but no cognitive abilities. As minimalists, we try not to "build in" the cognitive structures we are trying to explain, relying instead on entirely unsupervised learning algorithms to provide our robots with concepts that represent activities and objects. (All these algorithms find structure in multivariate time series, so they have been applied more generally to problems in temporal data mining.) Currently the robots are able to learn activities and activity boundaries; sensorimotor prototypes and primitive ontologies; and they have learned to identify words in speech and the denotations of the words. Even so, the robots' representations lack attributes that they will need to learn human language. Much of our current work is concerned with learning more compositional, denoting representations.
PAUL COHEN is Professor of Computer Science at the University of Massachusetts, Amherst. His PhD is from
Stanford University (1983) where he edited the Handbook of Artificial Intelligence with Avron Barr and Edward A.
Feigenbaum. Cohen's research focuses on learning and other statistical methods, planning, simulation, and the development
of robot minds. His publications and research projects are described at
http://eksl-www.cs.umass.edu/~cohen.
The Evening Lectures are free to both members and non-members of SGES.
For further information contact:
Dr. Hui Liu, Department of Computer Science, Birkbeck College hui@dcs.bbk.ac.uk