Comparison of three FPGA architectures for embedded multidimensional categorization through Kohonen's self-organizing maps

There is an increasing number of applications employing Self-Organizing Maps (SOM) for multidimensional categorization tasks. Those applications range from spoken- and written-word recognition and monitoring of industrial environments, to Internet of Things and image processing. SOM is a neural computing model with unsupervised learning, which is attractive for implementing embedded systems to operate autonomously in dynamic environments. Aiming to support comparisons among self-organizing computing methodologies, this work implements three different FPGA hardware architectures — distributed, centralized and hybrid — for executing SOM learning and recall phases. It also presents the contrasts between the characteristics of the elaborated circuits and the comparison of system design processes. Finally, comparisons with other published results of SOM neuromorphic systems are shown.

[1]  Andrew Hunter,et al.  A binary Self-Organizing Map and its FPGA implementation , 2009, 2009 International Joint Conference on Neural Networks.

[2]  Teuvo Kohonen,et al.  Self-organized formation of topologically correct feature maps , 2004, Biological Cybernetics.

[3]  Yutaka Maeda,et al.  Improved Learning Performance of Hardware Self-Organizing Map Using a Novel Neighborhood Function , 2015, IEEE Transactions on Neural Networks and Learning Systems.

[4]  Rafal Dlugosz,et al.  Kohonen winner takes all neural network realized on microcontrollers with AVR and ARM cores , 2010, Proceedings of the 17th International Conference Mixed Design of Integrated Circuits and Systems - MIXDES 2010.

[5]  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.

[6]  Hiroomi Hikawa,et al.  Supervised learning of DPLL based winner-take-all neural network , 2014, 2014 IEEE International Conference on Evolvable Systems.

[7]  Bertrand Granado,et al.  Toward a Sparse Self-Organizing Map for Neuromorphic Architectures , 2015, ACM J. Emerg. Technol. Comput. Syst..

[8]  Marta Kolasa,et al.  An FPGA implementation of the asynchronous programmable neighborhood mechanism for WTM Self-Organizing Map , 2011, Proceedings of the 18th International Conference Mixed Design of Integrated Circuits and Systems - MIXDES 2011.

[9]  Simon Haykin,et al.  Neural Networks and Learning Machines , 2010 .

[10]  Ulrich Rückert,et al.  IMPLEMENTATION OF SELF-ORGANIZING FEATURE MAPS IN RECONFIGURABLE HARDWARE , 2006 .

[11]  Rafael Gadea Gironés,et al.  A hardware design of a massive-parallel, modular NN-based vector quantizer for real-time video coding , 2008, Microprocess. Microsystems.

[12]  Jean-Charles Créput,et al.  Massively parallel cellular matrix model for self-organizing map applications , 2015, 2015 IEEE International Conference on Electronics, Circuits, and Systems (ICECS).

[13]  Timo Honkela,et al.  BIBLIOGRAPHY OF SELF-ORGANIZING MAP (SOM) PAPERS: 2002-2005 ADDENDUM , 2009 .