Multidimensional self organisation

Presents a technique that may be used for clustering in a very high dimensionality pattern space. The desirability of a self organising algorithm which can learn an internal representation for use in a pattern recogniser is shown. Using such an algorithm, subspace methods are brought together with an associative memory to form a pattern recogniser which employs unsupervised learning. The representation used for signal pattern clusters is based on topologically ordered units, each of which can label a complex area of pattern space. An adaption algorithm is given and shown to be insensitive to the variation in vector magnitudes which is found within a typical training set. A number of examples are given showing clustering of real grey scale, visual data and the reconstruction of exemplars using adaptive feedback. The application of this to vector quantisation and noise removal is demonstrated.<<ETX>>

[1]  Pentti Kanerva,et al.  Sparse Distributed Memory , 1988 .

[2]  Jorma Laaksonen,et al.  Variants of self-organizing maps , 1990, International 1989 Joint Conference on Neural Networks.

[3]  Yianni Attikiouzel,et al.  Kohonen's algorithm for the numerical parametrisation of manifolds , 1990, Pattern Recognit. Lett..

[4]  T. J. Stonham,et al.  Guide to pattern recognition using random-access memories , 1979 .

[5]  Teuvo Kohonen,et al.  Self-Organization and Associative Memory , 1988 .

[6]  S. P. Luttrell,et al.  Self-organisation: a derivation from first principles of a class of learning algorithms , 1989, International 1989 Joint Conference on Neural Networks.

[7]  Oliver G. Selfridge,et al.  Pattern recognition by machine , 1960 .

[8]  Teuvo Kohonen,et al.  The 'neural' phonetic typewriter , 1988, Computer.

[9]  C. H. Anderson.  A conditional probability interpretation of Kanerva's sparse distributed memory , 1989, International 1989 Joint Conference on Neural Networks.