Performance Comparison of SOM Based Hybrid Hardware Classifiers

This paper compares two hardware classifiers in various aspects. Both systems are hybrid network with SOM or Scalar SOM (SSOM) combined with Hebbian network. The SSOM is a simplified version of the SOM, which handles a single variable instead of vectors. The additional programmable network with the Hebbian learning capability, performs the category acquisition and naming. Two systems are described by VHDL and their classification performance as well as the circuit size and operating speed are compared. The results show that the SSOM based classifier exceeds the SOM based classifier in the circuit size and speed, while from the classification point of view, the performance of the SOM based classifier is better.