Definition
Connectionist Machine
A connectionist machine is a computational architecture consisting of a network of simple, non-linear processing elements. All world knowledge — both the program and the memory — is stored in the weighted connections between elements, not in the elements themselves.
Neural networks are connectionist machines. They stand in contrast to Von Neumann machines, where program and data reside in a separate memory and are fetched sequentially by a central processing unit.
Difference from a Von Neumann Machine
A Von Neumann machine separates processing from memory: a central processor fetches instructions and data from an addressable store. The program is a sequence of instructions.
A connectionist machine fuses processing and memory. The processing elements are many, simple, and non-linear. The program is the pattern of connections between them. Those same connections define the machine’s memory — there is no separate store.