![]() |
|
|||
![]() |
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
![]() |
|
![]() |
[an error occurred while processing this directive]
Another possible neural information representation is in frame language which could be again declarative or procedural. The basic description frame is declarative and embodies such information which cannot be reduced without losing the meaning of the process or events involved. On the other hand, there exists an octant frame which sheds off redundancy and provides minimal information for the neuronal control process. In a semiotic model, the frame language summarizes the languages of the relational types into multilevel hierarchical descriptions of the cellular complex. Such hierarchy is constituted by declarative information on the environment (descriptions of the physioanatomical aspects of the cells, axonic interconnections, etc.), and procedural aspects of the activities in the neural network (excitations and/or inhibitory biochemical responses, synaptic delays, etc.). It should be noted that this information is inherent in the associated memory of the network as beds and realizations mentioned earlier. 8.13 Informational Flow in the Neural Control ProcessIllustrated in Figure 8.6 is the flow chart of neural activity in the informational domains.
The functional aspects of the neural complex in the information plane as depicted by the semiotic model of Figure 8.6 can be summarized as follows: Pertinent to a set of input information that the neural complex receives from its environment, the information on the current output state of a neuron is fed back to the input of the network through an assembly of informational subsets. The first subset is a neural encoder (NC) which encodes the diverse information into syntagmatic chains. Such descriptions are fed to a neural analyzer (NA) which performs preliminary classification of the incoming information. If this information is needed for future use, it is stored in a knowledge-base (which is conventionally referred to as the memory (KB) of the neural complex) as traces. When the information in the neural analyzer is explicitly sufficient and cognizable, it is identified as information of the known category and directed to a correlator (NR) for the overlapping of the replicas, that is, for comparison with the trained patterns. Otherwise, it is considered as the information of unknown category and goes to a classifier (NL) for further categorization. The correlator serves to find a chain of control actions (to achieve the objective function) to match the controlling required. It also turns to the classifier to know the category of the information of the current neuronal state, if required. It may further use the data (memory) stored in the knowledge-base, as well. If the correlator finds a single control action, it informs the solver (NS) to let the relevant controlling activity to take place. If there are alternatives possible on the control strategy, the correlator may find it from the extrapolator (NE). The extrapolator seeks information from the knowledge-base pertinent to the output neuronal state and predicts the likely consequences of different control strategies. The responsibility of final controlling action rests with the solver. Thus, semiotic modeling of neural control processing permits the development of the entire control protocols in the information plane; and optimization of this semiotic control process provides a scope for the developments in neuroinformatics. Further, the semiotic approach in implementing the control information establishes its identity with the information existing at the input to the control. Thus, after being processed, the input information is transformed into control information which in turn provides input information about the neural state variables at the processor and, after the transit through the control means and the controlled system, input information about the functioning of the entire neural network. As dictated by the goal, each of the above stages is an information processing endeavor depicting sensing, recognition, prediction, decision making, and execution protocols. The pragmatics of such information and the characteristics of the processor can be evaluated as follows. Considering the role of information processing in the control of neural disorganization, the useful information (pragmatic assay) in reducing the disorganization in the sensing subgoal is given by: where The sensing subgoal refers to extracting only useful parts from the total information sufficient to impart knowledge to the controlled system. Similar to the above expression, the extent of useful information at the receiving synapse (recognizing the knowledge) can be written as: where ΔOD(R) refers to the reduction in disorganization with respect to recognition subgoal, Similar expressions can be written for prediction and decision-making information protocols aimed at minimizing the overall disorganization in the goal-related hierarchy.
Copyright © CRC Press LLC
![]() |
![]() |
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
![]() |
![]() |