Physically Evolving Networks (PENs) were first proposed by Alan Turing in his 1948 paper “Intelligent Machines” [1]. PENs capture some important features of the information processing of biological neural systems, such as mimicking animal and human error patterns, and implement massively parallel, non-algorithmic processes, where the sequence and concurrency of operations is determined at run time. PENs simulations on conventional digital computers have been successfully used for speech recognition, image analysis, and adaptive control. But PEN implementations on computers with von Neumann architecture are slow compared to other algorithms which are optimized for the sequential processing of explicit instructions. In the following we discuss hardware implementations of PENs. Would it be possible to fabricate a PEN of the size of a human brain with a billion times more neurons? And if so, how could such a PEN be programmed or trained?
[1]
JOHN F. Young.
Machine Intelligence
,
1971,
Nature.
[2]
Alfred Hubler,et al.
Hebbian learning in the agglomeration of conducting particles
,
1999
.
[3]
James P. Crutchfield,et al.
Order and disorder in open systems
,
2010,
Complex..
[4]
E. Capaldi,et al.
The organization of behavior.
,
1992,
Journal of applied behavior analysis.
[5]
Joseph Jun,et al.
Formation and structure of ramified charge transportation networks in an electromechanical system.
,
2005,
Proceedings of the National Academy of Sciences of the United States of America.
[6]
James P. Crutchfield,et al.
Order and disorder in open systems
,
2010
.