 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
Associative Memory At its simplest, an associative memory is a system which stores mappings from specific input representations to specific output representations. That is to say, a system that "associates" two patterns is one that, when presented only of othese patterns later, the other can be reliably recalled. Kohonen draws an analogy between associative memory and an adaptive filter function [2]. The filter can be viewed as taking an ordered set of input signals, and transforming them into another set of signals---the output of the filter. It is the notion of adaptation, allowing its internal structure to be altered by the transmitted signals, which introduces the concept of memory to the system. A further refinement in terminology is possible with regard to the associative memory concept, and is ubiquitous in connectionist (neural network) literature in particular. A memory that reproduces its input pattern as output is referred to as autoassociative (i.e. associating patterns with themselves). One that produces output patterns dissimilar to its inputs is termed heteroassociative (i.e. associating patterns with other patterns). Most associative memory implementations are realized as connectionist networks. Hopfield's collective computation network [1] serves as an excellent example of an autoassociative memory, whereas Rosenblatt's perceptron [3] is often utilized as a heteroassociator. There are many practical problems implementing effective associative memories however, most notably their inefficiency; the tendency is for them to fill up and become unreliable rather quickly. This is a long running open problem for both connectionism and adaptive filter theory---one that Kohonen refers to as the "problem of infinite state memory" [2]. References:
(Revsied November 2009) |
|
