... [O]ne of the most influential approaches to thinking about memory in recent years, known as connectionism, has abandoned the idea that a memory is an activated picture of a past event. Connectionist or neural network models are based on the principle that the brain stores engrams by increasing the strength of connections between different neurons that participate in encoding an experience. When we encode an experience, connections between active neurons become stronger, and this specific pattern of brain activity constitutes the engram. Later, as we try to remember the experience, a retrieval cue will induce another pattern of activity in the brain. If this pattern is similar enough to a previously encoded pattern, remembering will occur. The "memory" in a neural network model is not simply an activated engram, however. It is a unique pattern that emerges from the pooled contributions of the cue and the engram. A neural network combines information in the present environment with patterns that have been stored in the past, and the resulting mixture of the two is what the network remembers... When we remember, we complete a pattern with the best match available in memory; we do not shine a spotlight on a stored picture.
