ICANN ’93 pp 512-512 | Cite as

A Fuzzy Neural Architecture for Supervised Learning and Classification of Temporal Sequences

  • J. M. Cano Izquierdo
  • Y. A. Dimitriadis
  • J. López Coronado


The problem of learning and classifying ordered sequences is approached in this paper and a new architecture is proposed that incorporates elements from the Adaptive Resonance Theory (ART) and the theory of Fuzzy Sets.

The proposed architecture is composed of two stages. The first one is formed by a Fuzzy ART module that takes care of the unsupervised classification of the sequence components, and by a STORE module [1] that converts the sequence of category nodes provided by the Fuzzy ART to a spatial pattern. The second stage consists of another Fuzzy ART module and an associative memory.

The supervision is accomplished at the second stage, where a label is assigned to the global temporal sequence, that is introduced to the system. On the other hand, the spatial pattern produced by the first level is compared to the adaptive weights attached to the associative map, through a local distance that is based on the concept of fuzzy subset. This comparison permits the propagation of the supervision information to the first level, while taking advantage of the special form of the STORE patterns.

The proposed architecture extends the recently proposed Fuzzy ARTMAP [2] to the processing of temporal sequences, incorporating also an error checking mechanism at the level of an individual sequence component.

Experimental results are provided for the problem of on-line recognition of handwritten symbols, where each symbol is considered as a sequence of components. The obtained results compare favorably to those of a neural hierarchy [3] composed by ART2, ARTMAP and STORE modules, where the supervision information cannot be exploited at the symbol component level.


Associative Memory Fuzzy Subset Unsupervised Classification Adaptive Weight Adaptive Resonance Theory 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


  1. [1]
    G. Bradski, G.A. Carpenter and S. Grossberg, “Working memory networks for learning temporal order, with application to 3-D visual object recognition”, Neural Computation, vol. 4, pp. 270–276, 1992.CrossRefGoogle Scholar
  2. [2]
    G.A. Carpenter, S. Grossberg, N. Markuzon, J.H. Reynolds and D.B. Rosen, “Fuzzy ARTMAP: A neural network architecture for incremental supervised learning of analog multidimensional maps”, IEEE Trans. on Neural Networks, vol. 3, pp. 698–713, 1992.CrossRefGoogle Scholar
  3. [3]
    Y.A. Dimitriadis, J. Lopez Coronado, C. Garcia Moreno and J.M. Cano Izquierdo, “Online handwritten symbol recognition, using an ART based neural network hierarchy”, Proc. of the 1993 IEEE Conf. on Neural Networks, ICNN’93, vol. II, pp. 944–949, March 28–April 1, 1993, San Francisco.CrossRefGoogle Scholar

Copyright information

© Springer-Verlag London Limited 1993

Authors and Affiliations

  • J. M. Cano Izquierdo
    • 1
  • Y. A. Dimitriadis
    • 1
  • J. López Coronado
    • 1
  1. 1.Department of Automatic Control and Systems, School of Industrial EngineeringUniversity of ValladolidValladolidSpain

Personalised recommendations