Connectionism

Connectionism models cognition as the activity of networks of interconnected units whose changing connections encode information and support learning.

Artificial neural network: A computational system of interconnected units that transforms inputs through weighted connections. Connectionist models implement cognitive processes with networks of this general form.
Classical computationalism: The view that cognition operates through formal computations over symbolic representations. It contrasts with connectionist explanations based on distributed activity and learned weights.
Warren McCulloch: An American neurophysiologist who helped develop early mathematical models of neural activity. His work with Walter Pitts supplied a foundational formal model for connectionist networks.
Past-tense acquisition model: A connectionist model that learns English past-tense forms from examples, including regular and irregular verbs. It showed how networks might learn linguistic patterns without explicit grammatical rules.
Symbol grounding problem: The problem of explaining how symbols acquire meaning rather than merely referring to other symbols. Distributed, learned representations offer one proposed route from experience to meaningful concepts.
Distributed representation: A representation in which information is encoded across patterns of activity in multiple units. Connectionism explains concepts through shared patterns rather than single dedicated symbols.
Symbolic artificial intelligence: An approach to AI that represents knowledge with explicit symbols and manipulates them using rules. Its explicit symbols offer a different account of how intelligent behavior is produced.
Walter Pitts: An American logician whose mathematical work helped formalize neural computation. His collaboration with McCulloch shaped the early computational foundations of connectionism.
Rumelhart and McClelland's past-tense model: A 1986 neural-network model of how children learn English past-tense forms. The model sparked debate over whether connectionist learning could explain language development.
Compositionality: The property by which complex representations derive their meaning from parts and their arrangement. Connectionist models face the challenge of explaining systematic behavior with distributed representations.

Linked from 29 pages