|
 |
[an error occurred while processing this directive]
8.3 Information Base of Neurocybernetics
The neural complex handles information at three stages: The input stage (Figure 8.1), the processor stage, and the controlling stage (Figure 8.1). The input information includes all the objective knowledge on the structure, arrangement, and properties (such as the synaptic anatomy, physiology, and biochemical characteristics) of the participating neurons in the control endeavor. It also covers the details on the neural environment of the space (or domain) accommodating these neurons.
Figure 8.1 Information traffic in the neurocybernetic system S: Sensory input information; RE: Synaptic (recognition) information of the inputs; N: Intra- or extracellular disturbances (noise); OF: Axonal output information; DI: Decision-making information; OF: Learning-based objective function for self-control endeavors; PE: Predicted error information; EE: Error-correcting information for self-organization
The processing information characterizes the relevant properties and structure of the neural activity or state-transitional events across the interconnected neurons (these include the excitatory and inhibitory aspects, delays, threshold levels, inherent disturbances, and a variety of processing neuronal infrastructures). In essence, processing refers to all relevancies of the control center which strives to attain the objective function being fullfilled. It is a rule-based set of algorithms* which extracts useful information (from the inputs) and utilizes it to achieve the goal.
*An algorithm, in general, refers to a sequence of steps or codes or rules leading to specific results.
Once the information is processed, it represents a controlling information which is looped (forward or backward) into the system to refine the achievement of the goal or reduce the organizational deficiency that may prevail and offset the efforts in realizing the objective function. That is, the controlling information is the knowledge exerted to self-control (or regulate) the automata. The controlling information process is therefore a stand-alone strategy by itself, distinctly operating on its own as an adjunct to the main information processor.
The controlling information includes morsels of knowledge associated with sensing and recognition of the processed information (from the second stage), algorithmic manipulations on the sensed data for decision making, and strategies of predicting the errors. It supplies information to neural actuator(s) to execute the feedback or feed-forward controls through the organizational loop. The controlling information-processing works towards the realization of objectives or the goals with the minimization of errors.
Thus, the concept of information-processing in a neural complex viewed from a cybernetic angle assumes the informational structure is pertinent not only to perceiving knowledge from the source and analyzing it for its usefulness, but also to processing it (further) for applications towards achieving a self-organizing automaton. Specific to the application of information theory to neurocybernetics, the following general considerations are therefore implicit:
- In view of the automatic and adaptive control strategies involved in neurocybernetics, an informatic-based transfer function should be defined which refers to information-processing algorithms concerning the attainment of a target or objective function; and also in terms of informational efficiency functions they should assess how a given processing algorithm is realized by the control strategies involved (subject to constraints on speed of response, storage capacity, range of inputs, number of interconnections, number of learning iterations, etc.).
- There should be elucidation of methods by which the informational characteristics of the neural network can be derived by those strategies such as frequency domain analysis normally employed in the theory of adaptive controls.
- There should also be evaluation of the effectual or ineffectual aspects of classical stochastical theory of information in describing (either quantitatively or qualitatively) the control activities of a neural network. In general, statistical information describes the quantitative aspect of the message and does not portray the qualitative consideration of the utility of the message in contributing the attainment of the objective function. Further, the operation of a neural system is dynamic; and the statistical measure of information is unfortunately inadequate to include the nonstatistical part of the dynamic system such as its topological, combinatorial, and algorithmic features.
- The cybernetic perspective of a neural network represents the degree of organization of neural activity. Therefore, the corresponding informatic description of the neural system should not only address the memory considerations, but also enclave the control aspects of modeling and programming of the collective response of neurons.
- From a neurocybernetic viewpoint, the information theory should address the semiotic aspects of information, covering the syntatics which relate the formal properties of neurons and their combinations (variety) to the amount of information they carry; and semantics and pragmatics, which define the information content and the information utility of the neural signal elements (constituted by the binary state vectors of the cellular potential transitions), respectively.
- By enunciating a relation between the degree of self-organization versus the informatics of orderliness, a new approach as applicable to neural cybernetics can be conceived. Wiener indicated such a trend as to incorporate and extend information theory to neural cybernetics via semiotic considerations.
- The concept of neurocybernetic informational theory rests upon the threshold of distinguishability of neural state variables and the amount of their variety pertinent to the self-control process.
- Analysis and synthesis of man-machine systems as done in the development of computer architecture and artificial neural networks mimicking the biological neurons refer to modeling and programming only at the information structural level. Such modeling and programming via information theory in the neurocybernetic domain should, however, broadly refer to:
- a. Informational description of the global neural complex.
- b. Similarity or dissimilarity criteria with regard to the objective structures, information structures, and information flows specified in terms of entropy parameters of the neural complex.
- c. Establishing a similarity between informational functions of processing by self-organizing (control) centers of the interconnected cells.
|