Subsymbolic computation

Computation over quantities such as distributed activation patterns and learned weights, without requiring every meaningful entity to occupy a separate explicit token. Neural networks are common examples. A distributed implementation can nevertheless learn or implement structured operations.