|
|
|
|
|
|
|
duced. As an example (simplified for brevity) the sensor-integrator-receptor-producer complex  maintains production of X2P if a metabolite with initial radical X1 ever makes an appearance. In fact, if sensors and receptors can have the same range of sensitivity as the arguments of broadcast units, it is easy to show that there is an appropriate Britten-Davidson model for producing any arbitrarily given sequence of products. |
|
|
|
|
|
|
|
|
A very similar representation can be produced for the lymphocyte immune network, well described by Niels K. Jerne (1973) and presented more technically by M. Sela (1973). In this case the "detectors" are combining sites on antibody molecules produced by lymphocyte cells. The environmental "signals" are invading antigens (e.g., foreign protein molecules). The presence of a detected antigen causes the production of additional lymphocytes (additional "broadcast units," see usage (7) of section 3) which in turn secrete additional antibodies which combine with (and neutralize) the antigens. |
|
|
|
|
|
|
|
|
A bit further afield the broadcast language can also serve for a straightforward representation of the cell assembly model of the central nervous system (section 3.6). Here the broadcast units are cell assemblies while the "to-whom-it-may-concern" aspect of the broadcast language is reasonably approximated by the large number of neurons (103 to 104) in other assemblies contacted by each neuron in a given cell assembly. (More specific interconnections can be represented by appropriate "tagging" (prefixes) as in section 3.) Then, synaptic "learning" rules which induce fractionation and recruitment in cell assemblies find counterparts in generalized genetic operators which modify representations. Closely associated cell assemblies become the counterparts of tested representational components (cf. schemata), and so on. (The interested reader should consult Plum [1972] for the details of a related model.) |
|
|
|
|
|
|
|
|
In the context of the broadcast language, the cell assembly model fits smoothly with the predictive modeling technique of section 3.4. A discussion of the latter implementation also gives an indication of how the broadcast language is applied to artificial systems. One implementation which emphasizes the cell-assembly/predictive-modeling fit relies on a set of behavioral units which generate action sequences and are modified on the basis of the outcome. Each behavioral unit consists of a population of behavioral atoms realized as devices in the broadcast language. If we look back to the search strategies of Figure 6 it is the detectors which have a role comparable to the atoms here. In the broadcast language, the detectors would be broadcast units (or sets of them) with arguments corresponding to the conditions defining the detector. (For example, the atom corresponding to d1 in Figure 6 would be activated by any 4-by-4 array with 8 or more dark squares.) |
|
|
|
|
|