SELFCONSTRICTION: A smart trade-off?

mervyn at mervyn at
Tue Mar 25 15:44:00 EST 1997

The following paper is available as html-document (english and dutch
version online). The paper also provides links to references and
general neuroscience resources. All comment and reactions welcome!
In this article the imagined evolution of an organism introduces a
controller that exhibits simple behavior by means of a few neural
connections. Step by step this controller is developed until a model
occurs that shows predictive power for certain aspects of human cognition
and behavior. This model could explain the existence of the trade-off
between optimal adaption to changing a context and optimal performance
within a stable context. It is suggested that the hypothalamus and the
four diffuse systems are responsible for this trade-off. Computer
simulations that illustrate the features of the developed
'self-constricting' neural network are reported on.
In this paragraph a imaginary organism is discussed that can search for
food by means of a few neurons. A choice is made between two equally
competitive controller variants is made. A choice that later on in the
evolution of the organism will prove to be decisive for its developing
mental abilities.
In this paragraph the symmetrical variant is interpreted as being a
controller element that can match a measured aspect of the organism
physical state to a target state (e.g. a full stomach). By means of an
array of these so called mismatch detectors a neural network can be
trained to help satisfy the needs of more complex organism.
In this paragraph the organism is developed further: external sensors
provide it with information coming from its environment. The
'self-constricting' property of the symmetrical variant turns out to be
able to force the neural network to define, for each moment, a target
state for external sensors on the basis of earlier experience. The more
complex this experience becomes, the more complex the behavior will it
exhibit to avoid re-adjustment of its world model.
In this final paragraph a process is discussed that, thanks to the used
reinforcement mechanism, is able to tackle the settling-time problem. The
process (network-modulation), its distinctive side-effects included, is
proposed to occur within the human brain as well, in which it is
'implemented' by the hypothalamus the four diffuse systems.

More information about the Neur-sci mailing list