An assocative memory is one which tries to learn relationshop between an imput and an output so that when given an unseen input it can allocate an appropriate output. In some case the output is very different form the output. For example, where an image is associated with a class. In others the memory can be used as an auto-encoder where the input and output are identical -- this form of auto-associative memory initially does not seem useful, however it may be used where the input is in some way incomplete or noisy and the memory recreates a full or noise-free image.
Used on page 117