SANS annual report 1994
Previous 
Up 
Next
Artificial Neural Networks
Artificial Neural Networks
Our long-term goal for investigations of neural computation is the
design of a general purpose adaptive control system for technical
applications in which a high degree of autonomous operation is a key
feature. Important functional characteristics include real-time,
spatio-temporal pattern processing capabilities for sensory,
perceptual, and motor control purposes, noise tolerant learning, and
efficient goal-direction. A massively parallel and distributed
architecture, and tolerance to hardware faults are other important
features. The idea is to engage in both basic and applied research
which makes possible a fruitful interaction between theory and
practice. The models are evaluated in a set of test- and pilot
applications.
We are convinced that drawing on analogies with neural information
processing is an essential component in researching the technical
aspects of ANN:s. This applies at the problem level (demands on
autonomous systems etc.) as well as at the system's architecture level
(necessary building blocks, types of interactions etc.) and the
component level (neurons and synapses). Apart from this, neural
computation also has important ties to several other disciplines,
e.g. statistical pattern recognition and inference methods, adaptive
control, and information theory.
Subtopics
Next