Implementation Issues of Kohonen Self-Organizing Map Realized on FPGA

Presented are the investigations showing an impact of the length of data signals in hardware implemented Kohonen Self-Organizing Maps (SOM) on the quality of the learning process. The aim of this work was to determine the allowable reduction of the number of bits in particu- lar signals that does not deteriorate the network behavior. The eciency of the learning process has been quantied by using the quantization error. The results obtained for the SOM realized on Field Programmable Gate Array (FPGA), as well as by means of the software model of the SOM show that the smallest allowable resolution (expressed in bits) of the weight sig- nals equals seven, while the minimal bit length of the neighborhood signal ranges from 3 to 6 (depending on the map topology). For such values and properly selected values of other parameters the learning process remains undisturbed. Reducing the number of bits has an inuence on the number of neurons that can be synthesized on a single FPGA device.

[1]  Michel Verleysen,et al.  Self-organizing maps with recursive neighborhood adaptation , 2002, Neural Networks.

[2]  Hiroomi Hikawa.  FPGA implementation of self organizing map with digital phase locked loops , 2005, Neural Networks.

[3]  Witold Pedrycz,et al.  Programmable triangular neighborhood function for kohonen self-organizing map implemented on chip , 2010, Proceedings of the 17th International Conference Mixed Design of Integrated Circuits and Systems - MIXDES 2010.

[4]  Jorge Peña,et al.  Digital Hardware Architectures of Kohonen's Self Organizing Feature Maps with Exponential Neighboring Function , 2006, 2006 IEEE International Conference on Reconfigurable Computing and FPGA's (ReConFig 2006).

[5]  W. Kurdthongmee,et al.  A novel hardware-oriented Kohonen SOM image compression algorithm and its FPGA implementation , 2008, J. Syst. Archit..