Parallel distributed computing within interconnected nets.
A system also can be understood as a net of simple components with propagation of flows.
E. ANDREEWSKY explains: "According to Connectionism, the brain and its neurons explain behavior, and the terms used in this field, such as neurons, synapses, activation, propagation, suggest a biological explanation for cognitive phenomena. The former A.I. abstract symbolic representations become a set of very simple elements (formal neurons) functioning in parallel. Like the perceptron, connectionist systems work through learning & Connectionist systems do not have central units, and each of their elements functions in its local environment in order to determine a global cooperation, that is the system's organization. The network shapes itself progressively with the stimuli it receives; in other words, its learning activity, which is continuous during its implementation, results in changes of the network itself (whereas classically, learning is only change in knowledge representation)" (1993, p.192-193).
Of course "change in knowledge representation" implies the previous existence of knowledge. Thus such a view of learning does not explain the genesis, or better, the autogenesis of knowledge.
The physio-psychological version of connectionism is an extension of D. RUMELHART, J. McCLELLAND and G. HINTON's Parallel Distributed Processing, which is a model for information processing.
As to the mechanism of connectionist construction, E. ANDREEWSKY resumes J. Mc CLELLAND and D. RUMELHART explanations: "The communication between the different units of this system no longer takes place by means of messages as in classical Artificial Intelligence, but through activation values: numbers and not symbols. The interpretation of the operation is no longer obtained in terms of messages transmited between modules of the system; it represents the activated states of the whole system" (p.193).
Morevover, and in accordance with D. HEBB's reinforcement through repeated stimulation, connectionism implies reinforcement of pathways by frequent use.
Connectionism is also obviously and closely related to the more general problem of autogenesis in living and in social systems.
Connectionist network: "A parallel-processing system made of many simple computational units, linked (as brain-cells are) by excitatory or inhibitory connections" (M. BODEN, 1990, p.124).
This author explains: "One unit modifies another's activity to different degrees, dependent on the relevant connection-weight (expressed as a number between plus-one and minusone). The details of these weight changes are governed by differential equations, like those used in physics. A concept is represented as a "stable activity-pattern across the entire system".
"In networks that can learn, the connection weights are continually adjusted to maximize the probability of reaching equilibrium. Connections used often are strengthened, and if two units are activated simultaneously, then connection weights are adjusted to make this more likely in the future" (Ibid).
Moreover: "The larger the network, and the more distinct the patterns, the more associations can be learnt" (Ibid).
- 1) General information
- 2) Methodology or model
- 3) Epistemology, ontology and semantics
- 4) Human sciences
- 5) Discipline oriented
To cite this page, please use the following information:
Bertalanffy Center for the Study of Systems Science (2020). Title of the entry. In Charles François (Ed.), International Encyclopedia of Systems and Cybernetics (2). Retrieved from www.systemspedia.org/[full/url]
We thank the following partners for making the open access of this volume possible: