scientific article; zbMATH DE number 4085057
A fragment model of associative memory of a neural network defined by \textit{G. Palm} [Biol. Cybern. 36, 19-31 (1980; Zbl 0426.68102)] is introduced. The task of the fragment is to store and reproduce a mapping of a set of t queries onto a set of t answers. Queries and answers are coded by boolean n-ary and m-ary vectors, respectively. A query code of length n for storage capacity t is given by a boolean matrix X of size \(n\times t\) with code vectors as columns; an answer code is given in an analogous way. An ordered pair (X,Y) of codes has the property of stored information reproduction if any mapping p: \(X\to Y\) based on a permutation of \(\{\) 1,2,...,t\(\}\) is representable by \(y_{p(u)}=A^ p x_ u\), \(u=1,2,...,t\), where \(A^ p\) is an associative memory matrix for the mapping p. A pair (X,Y) has the reproduction property iff X is a disjunctive (s,t)-code, where s is the maximum number of ``1 in a row of Y. In the paper formulas giving interrelations of parameters m, n, t necessary for the reproduction property of codes are summarized.
- Context-dependent associations in linear distributed memories
- On optimal nonlinear associative recall
- An operational happens-before memory model
- Adding a bias to vector models of association memory provides item memory for free
- Foundations of the neural network assembly memory model
- About the theory of mnemofunctions and its
- A memory model with novel behaviour in sequential learning
- Superimposed codes and threshold group testing
- Minimum number of input clues in robust information retrieval
- Mathematical models of human memory
- A cache for an associative memory
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3814833)