next up previous
Next: References Up: A Basis for Long-Range Previous: Attention

Conclusions

In the earlier part of this paper, we have explored, both mathematically and by simulation, various features of the TH-NRT-C system modelled as a purely inhibitory net for NRT fed excitatorily from the Thalamus and feeding back disinhibitorily to the thalamus, which itself acts as an excitatory feedback net to both the NRT and the thalamus. The mathematical analysis show that the TH-NRT-C system can act both in an enhancement manner to achieve competition between local inputs and also in a global competitive mode. The manner in which these appear to be most effective in terms of the fan-in and fan-out were discussed in a simplified (linearised) model. Sensitivity to the inputs of the global competitive mode, in particular to the form of dendro-dendritic coupling on the NRT, was also explored, and a particular form of this coupling found to give similar sensitivity to that arising in wave-structures in physical systems far from equilibrium [13], being amenable both to mathematical analysis and simulation. This particular d-d synaptic functionality seems preferred, from the point of view of sensitivity to inputs, compared to other possibilities; modelling the actual NRT would require a more detailed analysis of d-d synaptic action to guarantee this is correct.

Besides considering on-going activity, learning effects were considered in Taylor & Alavi [48], again by mathematical analysis and by simulation. The latter gave a periodic structure to afferent and lateral cortical connection weights. Neurophysiological evidence in support of such a global wave structure of cortical storage was presented, although conjoint effects of other storage mechanisms [9,35] would need a more careful analysis to disentagle.

One of the main purposes of this study was to describe a unique system in the brain (the TH-NRT-C complex) and the possibility for it to act as a global control mechanism. The manner in which this does so for allowing efficient filtering of multiple sets of on-going cortical activity (in different modalities) is explored elsewhere [45,46].

In the second half of the paper, a further sub-cortical system of lateral inhibitory cells was considered as a supporting system for further long-range competitive effects in cortex, how more specifically heteromodal. The ACTION network, as a simplified model of such a system, has been proposed here to allow for the support of active memory, transformation of population vectors, temporal sequence storage, comparator action and attention. The main structure of the model was presented in the ACTION Network section, and in particular its neurobiological underpinning emphasised. Its support of active memory was demonstrated in the Active Memory section, and in particular the flip-flop aspect of its activity, with many possible attractors, indicated. It was also demonstrated how the development of the population vector, suitably encoded, explains the observations in [20] on the growth of this vector during a delay period. Finally, in the ACTION Network section, the manner in which a competition between two different cues can be supported by the ACTION net was demonstrated, as the possible basis for the attentional effects on IT cells presented in [8]. Delayed memory cells in IT or the parietal-temporal-occipital region, being heteromodal cortex, would be expected to have similar connections to the basal ganglia as frontal lobe. However, the presence of flip-flops in such posterior circuits needs to be investigated more fully.


next up previous
Next: References Up: A Basis for Long-Range Previous: Attention