- A network
- A vector of activations (length n)
- Sometimes a separate vector of outputs and/or phase angles
- A matrix of weights (dimensions n X n)
- A task
- A set of input vectors
- (Sometimes) a set of target vectors
- An activation rule
- A weight rule
- Connectionist models implement parallel constraint satisfaction in a straightforward way. This permits processing as the interaction of diverse knowledge sources.
- Connectionist networks are good at solving best-match problems.
- When a familiar pattern enters the memory system, the response is a stronger version of the input (recognition).
- When an unfamiliar pattern enters the memory system, it is dampened (unfamiliarity).
- When part of a familiar pattern enters the memory system, the system fills in the missing parts (recall, content-addressable memory).
- When a pattern similar to a stored pattern enters the memory system, the response is a version of the input distorted toward the stored pattern (assimilation).
- When a number of similar patterns have been stored, the system will respond to the central tendency of the stored patterns, even if the central tendency itself never appeared (prototype effects).
- Connectionist networks are robust.
- They are resistant to noise.
- They gracefully degrade when they are damaged.
- They gracefully degrade when they are overloaded with information.
- Connectionist networks are based loosely on brains.
- Control is distributed.
- Computation is handled by massively connected, simple processors which repeatedly update on the basis of weighted inputs.
- Learning consists in enhancing or inhibiting on the basis of local information the degree to which signals are transmitted between processors.
- Memories are distributed.
- (Distributed) connectionist representations are flexible.
- They are graded.
- They are context-dependent.
- Connectionist models offer the possibility of grounding
internal representations in perception, action, and affect.
- Low-level representations are continuous.
- High-level representations can be categorical.
- Similarity between patterns is represented directly, and there is automatic generalization to similar inputs.
- There is a wide range of relatively simple, local connectionist learning algorithms.
- Variable binding
- Recursion
- Reflection
- Structured representations
- Multiple instantiations