Complex systems and Biological physics seminar [before December 2013]

Transparent neural networks: a cognitive architecture for automatic development of intelligence

by Claes Strannegård (Göteborg University)

Europe/Stockholm
Description

I present the transparent neural networks, a cognitive architecture that integrates symbolic and sub-symbolic reasoning in a monolithic computational framework.

The ultimate goal is a computational model capable of developing intelligence in a completely autonomous fashion, solely by interacting with the environment.

Our present model perceives sensory data from the surrounding world and gradually develops intelligence by observing, classifying, and reasoning about the environment. Memory structures are added and removed dynamically. Thus we obtain a model capable of automatic concept formation as well as deductive and inductive reasoning.