|
The human brain is autonomously active. Understanding
the functional role of this self-sustained neural activity,
and its interplay with the sensory data input stream, is a
central quest in cognitive system research. We present first a
short overview of both recent experimental results and several
distinct theory approaches regarding the modelling of neural
networks showing a non-trivial eigendynamics in terms of
transient states.
We then discuss our approach based on object encoding in terms of specific local neural ensembles, the cliques of graph theory, using a single guiding principle, neural competition, for the formulation of the neural dynamics. We implement this guiding principle both for the internal transient-state dynamics as well as for the interaction of the internal neural activity with the incoming sensory stimuli. Unsupervised and online Hebbian-style local synaptic plasticities are the activated by the sensory stimuli, leading to an emergent cognitive capability. The system performs, With no explicit algorithmic encoding, a non-linear independent component analysis of the sensory input stream and semantic learning of the internal neural dynamics occurs. |
|