Abstract
This paper introduces a new neural network training algorithm, Hypercube Separation (HCS) algorithm which is very fast and guaranteed to learn. HCS is a simple algorithm suitable for hardware implementation which classifies different input patterns presented to it through the formation of multiple hyperplanes. The performance of the HCS algorithm is compared to that of the Binary Synaptic Weights (BSW) algorithm and to the Backpropagation (BP) algorithm in solving the two spiral problem, which is an almost pathological problem for pattern separation. The HCS algorithm was able to successfully separate the input patterns, requiring three orders of magnitude less training time than the BP algorithm and one order of magnitude less hidden layer nodes than the BSW algorithm. We also present the application of HCS to on-line handwritten character recognition with good results, especially when the simple nature of the algorithm is taken into consideration.
Similar content being viewed by others
References
H.M.A. Andree et al., “A comparison study of binary feedforward neural networks and digital circuits,” Neural Networks, vol. 6, pp. 785–790, 1993.
J. Dayhoff, “Neural network architectures: An introduction,” Van Nostrand Reinhold, New York, 1990.
J. Guo et al., “Recognition of handwritten characters using pattern transformation method with cosine function,” Trans. of the IEICE, vol. J76-D-II, no. 4, pp. 835–842, 1993 (in Japanese).
D. Hammerstrom, “Working with neural networks,” IEEE Spectrum, vol. 30, no. 7, pp. 46–53, 1993.
K. Imai, K. Gouhara, and Y. Uchikawa, “Handwritten letter recognition for noisy images using the three-layered BP-models,” Trans. of the IEICE, vol. J75-D-II, no. 2, pp. 314–323, 1992 (in Japanese).
T. Kawatani and N. Miyamoto, “shape quantification of handprinted numerals using recognition results and its application,” Trans. of the IEICE, vol. J74-D-II, no. 3, pp. 357–365, 1991 (in Japanese).
S. Kurakake and K. Ishi, “Evaluation of handprinted character recognition systems using distorted characters,” Trans. of the IEICE, vol. J74-D-II, no. 9, pp. 1208–1216, 1991 (in Japanese).
Y. Maeda, H. Yamashita, and Y. Kanata, “Learning rules for multilayer neural networks using a difference approximation,” Proceedings of the IJCNN'91, Singapore, 1991, pp. 628–633.
J. L. McClelland, D.E. Rumelhart et al., Parallel Distributed Processing, The MIT Press: Cambridge, Massachusetts, 1986.
M. Mezard and J.P. Nadal, “Learning in feedforward layered neworks: The Tiling Algorithm,” J. Phys., A: Math. Gen., vol. 22, pp. 2191–2203, 1989.
S. Mori et al., “Research on machine recognition of handprinted characters,” IEEE Trans. on Pat. An. and Machine Int., vol. PAMI-6, no. 4, pp. 386–405, 1984.
T. Ohtomo, T. Hara, and K. Hara, “Two-stage recognition method of handwritten chinese characters using an integrated neural network model,” Trans. of the IEICE, vol. J74-D-II, no. 2, pp. 158–165, 1991 (in Japanese).
T.J. Sejnowski and C.R. Rosenberg, “NETtalk: A parallel network that learns to read aloud,” John Hopkins University, Tech. Rep. JHU/EECS-86-01,1986.
W. Suewatanakul and D.M. Himmelblau, “Comparison of artificial neural networks and traditional classifiers via the two-spiral problem,” in Proc. of SIMTEC'92, 1992, pp. 431–438.
F. Ulgen and N. Akamatsu, “A fast algorithm with a guarantee to learn: Binary synaptic weights algorithm on neural networks,” in Proc. of SIMTEC'92, 1992, pp. 469–474.
K. Yamada, J. Tsukumo, and T. Temma, “Character recognition by neural networks with standstill evading learning algorithm,” Trans. of the IEICE, vol. J75-D-II, no. 11, pp. 1798–1808, 1992 (in Japanese).
Author information
Authors and Affiliations
Rights and permissions
About this article
Cite this article
Ulgen, F., Akamatsu, N. & Iwasa, T. The Hypercube Separation algorithm: A fast and efficient algorithm for on-line handwritten character recognition. Appl Intell 6, 101–116 (1996). https://doi.org/10.1007/BF00117811
Received:
Revised:
Accepted:
Issue Date:
DOI: https://doi.org/10.1007/BF00117811