Autoassociative memory

Autoassociative memory, also known as auto-association memory or an autoassociation network, is any type of memory that is able to retrieve a piece of data from only a tiny sample of itself. They are very effective in de-noising or removing interference from the input and can be used to determine whether the given input is “known” or “unknown”.

In artificial neural networks, examples include variational autoencoder, denoising autoencoder, Hopfield network.

In reference to computer memory, the idea of associative memory is also referred to as Content-addressable memory (CAM).

The net is said to recognize a “known” vector if the net produces a pattern of activation on the output units which is the same as one of the vectors stored in it.

Background

Standard memories (data storage) are organized by being indexed by positional memory addresses which are also used for data retrieval.

Autoassociative memories are organized in such a way that data is stored in a graph like system with connection weights based on the number of inherent associative connections between two memories which makes it possible to query it using a memory already contained in the associative memory as query-key and retrieve that memory and closely connected memories at the same time. Hopfield networks have been shown to act as autoassociative memory since they are capable of remembering data by observing a portion of that data.

Iterative Autoassociative Net

In some cases, an auto-associative net does not reproduce a stored pattern the first time around, but if the result of the first showing is input to the net again, the stored pattern is reproduced. They are three further kinds: Recurrent linear auto-associator, Brain-State-in-a-Box net, and Discrete Hopfield net.