Monthly
288 pp. per issue
6 x 9, illustrated
ISSN
0899-7667
E-ISSN
1530-888X
2014 Impact factor:
2.21

Neural Computation

November 1992, Vol. 4, No. 6, Pages 880-887
(doi: 10.1162/neco.1992.4.6.880)
© 1992 Massachusetts Institute of Technology
How to Incorporate New Pattern Pairs without Having to Teach the Previously Acquired Pattern Pairs
Article PDF (325.62 KB)
Abstract

In this paper, we deal with the problem of associative memory synthesis. The particular issue that we wish to explore is the ability to store new input-output pattern pairs without having to modify the path weights corresponding to the already taught pattern pairs. The approach to the solution of this problem is via interpolation carried out in a Reproducing Kernel Hilbert Space setting. An orthogonalization procedure carried out on a properly chosen set of functions leads to the solution of the problem.