Abstract
In this paper, we propose a new type of information-theoretic method called “input information maximization” to improve the performance of self-organizing maps. We consider outputs from input neurons by focusing on winning neurons. The outputs are based on the difference between input neurons and the corresponding winning neurons. Then, we compute the uncertainty of input neurons by normalizing the outputs. Input information is defined as a decrease in the uncertainty of input neurons from a maximum and observed value. When input information increases, fewer input neurons tend to be activated. In the maximum state, only one neuron is on, and all others are off. We applied the method to two data sets, namely, the Senate and voting attitude data sets. In both, experimental results confirmed that when input information increased, quantization and topographic errors decreased. In addition, clearer class structure could be extracted by increasing input information. In comparison to our previous methods to detect the importance of input neurons, the present method turned out to be good at producing faithful representations with much more simplified computational procedures.
Similar content being viewed by others
References
Kohonen T (1988) Self-organization and associative memory. Springer, New York
Kohonen T (1995) Self-organizing maps. Springer, Berlin
Kohonen T (1990) The self-organization map. In: Proceedings of the IEEE, vol 78, pp 1464–1480
Linsker R (1988) Self-organization in a perceptual network. Computer 21:105–117
Linsker R (1989) How to generate ordered maps by maximizing the mutual information between input and output. Neural Comput 1:402–411
Linsker R (1992) Local synaptic rules suffice to maximize mutual information in a linear network. Neural Comput 4:691–702
Linsker R (2005) Improved local learning rule for information maximization and related applications. Neural Netw 18:261–265
Sindhwani V, Raskshit S, Deodhare D, Erdogmus D, Principe JC (2004) Feature selection in mlps and svms based on maximum output information. IEEE Trans Neural Netw 15(4):937–948
Fisher JW, Principe JC (1998) A methodology for information theoretic feature extraction. In: IEEE world congress on computational intelligence, pp 1712–1716
Morejon RA, Principe JC (2005) A new classifier based information-theoretic learning with unlabel data. Neural Netw 18:719–726
Jeon K, Xu J, Erdogmus D, Principe JC (2005) A new classifier based information-theoretic learning with unlabel data. Neural Netw 18:719–726
Lehn-Schioler DET, Hegde A, Principe JC (2004) Vector-quantization using information theoretic concepts. Nat Comput 4(1):39–51
Torkkola K (2001) Nonlinear feature transform using maximum mutual information. In: Proceedings of international joint conference on neural networks, pp 2756–2761
Torkkola K (2003) Feature extraction by non-parametric mutual information maximization. J Mach Learn Res 3:1415–1438
Hammer B, Hasenfuss A, Villmann T (2007) Magnification control for batch neural gas. Neural Comput 70:1225–1234
Hasenfuss A, Hammer B, Geweniger T, Willmann T (2008) Magnification control in relative neural gas. In: Proceedings of ESANN2008
Villmann T, Merenyi E, Hammer B (2003) Neural maps in remote sensing image analysis. Neural Netw 16:389–403
Villmann T, Claussen JC (2006) Magnification control in self-organizing maps and neural gas. Neural Comput 18:446–469
Merenyi E, Jain A, Villmann T (2007) Explicit magnification control of self-organizing maps for forbidden data. IEEE Trans Neural Netw 18(3):786–797
Van Hulle M (1996) Topographic map formation by maximizing unconditional entropy: a plausible strategy for ‘on-line’ unsupervised competitive learning and nonparametric density estimation. IEEE Trans Neural Netw 7(5):1299–1305
Hulle MMV (1997) The formation of topographic maps that maximize the average mutual information of the output responses to noiseless input signals. Neural Comput 9(3):595–606
Van Hulle MM (1999) Faithful representations with topographic maps. Neural Netw 12(6):803–823
Van Hulle MM (2004) Entropy-based kernel modeling for topographic map formation. IEEE Trans Neural Netw 15(4):850–858
Kamimura R, Kamimura T, Shultz TR (2001) Information theoretic competitive learning and linguistic rule acquisition. Trans Jpn Soc Artif Intell 16(2):287–298
Kamimura R, Kamimura T, Uchida O (2001) Flexible feature discovery and structural information control. Connect Sci 13(4):323–347
Kamimura R (2003) Information-theoretic competitive learning with inverse Euclidean distance output units. Neural Process Lett 18:163–184
Rumelhart DE, Zipser D (1986) Feature discovery by competitive learning. In: Rumelhart DE, et al (eds) Parallel distributed processing, vol 1. MIT Press, Cambridge, pp 151–193
Grossberg S (1987) Competitive learning: from interactive activation to adaptive resonance. Cogn Sci 11:23–63
Fukushima K (1975) Cognitron: a self-organizing multi-layered neural network. Biol Cybern 20:121–136
Fukushima K (1975) Neocognitron: a hierarchical neural network capable of visual pattern recognition. Biol Cybern 20:121–136
Fukushima K (1983) Neocognitron: a neural network model for a mechanism of visual pattern recognition. IEEE Trans Syst Man Cybern 13:826–834
DeSieno D (1988) Adding a conscience to competitive learning. In: Proceedings of IEEE international conference on neural networks, San Diego. IEEE, New York, pp 117–124
Ahalt SC, Krishnamurthy AK, Chen P, Melton DE (1990) Competitive learning algorithms for vector quantization. Neural Netw 3:277–290
Xu L (1993) Rival penalized competitive learning for clustering analysis, RBF net, and curve detection. IEEE Trans Neural Netw 4(4):636–649
Luk A, Lien S (2000) Properties of the generalized lotto-type competitive learning. In: Proceedings of international conference on neural information processing. Morgan Kaufmann, San Mateo, pp 1180–1185
Katsavounidis I, Kuo CCJ, Zhang Z. A new initialization technique for generalized Lloyd iteration. IEEE Signal Process Lett 1
Xiong H, Swamy MNS, Ahmad MO (2004) Competitive splitting for codebook initialization. IEEE Signal Process Lett 11:474–477
Xiong H, Swamy MNS (2004) Branching competitive learning network: a novel self-creating model. IEEE Trans Neural Netw 15(2):417–429
Wu S, Liew W, Yuan H, Yang M (2004) Cluster analysis of gene expression data based on self-splitting and merging competitive learning. IEEE Trans Inf Technol Biomed 8(1):5–15
Maass W (2000) On the computational power of winner-take-all. Neural Comput 12(11):2519–2535
Wolfe W, Mathis D, Anderson C, Rothman J, Gottler M, Brady G, Walker R, Duane G, Alaghband G (1991) K-winner networks. IEEE Trans Neural Netw 2(2):310–315
Urahama K, Nagao T (1995) K-winners-take-all circuit with o (n) complexity. IEEE Trans Neural Netw 6(3):776–778
Ridella S, Rovetta S, Zunino R (2001) K-winner machines for pattern classification. IEEE Trans Neural Netw 12(2):371–385
Yen JC, Guo JI, Chen HC (1998) A new k-winners-take-all neural networks and its array architecture. IEEE Trans Neural Netw 9(5):901–912
Calvert BD, Marinov GA (2000) Another k-winners-take-all analog neural network. IEEE Trans Neural Netw 11(4):829–838
Martinez TM, Berkovich SG, Schulten KJ (1993) Neural-gas network for vector quantization and its application to time-series prediction. IEEE Trans Neural Netw 4(4):558–569
Zeger EYK, Gersho A (1992) Competitive learning and soft competition for vector quantizer designs. IEEE Trans Signal Process 40(2):294–308
Kamimura R (2008) Conditional information and information loss for flexible feature extraction. In: IEEE international joint conference on neural networks, IJCNN 2008. IEEE, New York, pp 2074–2083
Kamimura R (2007) Information loss to extract distinctive features in competitive learning. In: IEEE international conference on systems, man and cybernetics, ISIC 2007. IEEE, New York, pp 1217–1222
Kamimura R (2010) Information-theoretic enhancement learning and its application to visualization of self-organizing maps. Neurocomputing 73(13–15):2642–2664
Kamimura R (2011) Double enhancement learning for explicit internal representations: unifying self-enhancement and information enhancement to incorporate information on input variables. Appl Intell, 1–23
Kamimura R (2011) Self-enhancement learning: target-creating learning and its application to self-organizing maps. Biol Cybern, 1–34
Kamimura R (2011) Selective information enhancement learning for creating interpretable representations in competitive learning. Neural Netw 24(4):387–405
Shie J-D, Chen S-M (2008) Feature subset selection based on fuzzy entropy measures for handling classification problems. Appl Intell 28(1):69–82
Shen Q, Jensen R (2008) Approximation-based feature selection and application for algae population estimation. Appl Intell 28(2):167–181
Chen C-M, Lee H-M, Hwang C-W (2005) A hierarchical neural network document classifier with linguistic feature selection. Appl Intell 23(3):277–294
Lee S, Park Y-T, d’Auriol BJ et al (2012) A novel feature selection method based on normalized mutual information. Appl Intell 37(1):100–120
Ye H, Lo BW (2000) Feature competitive algorithm for dimension reduction of the self-organizing map input space. Appl Intell 13(3):215–230
Vesanto J, Himberg J, Alhoniemi E, Parhankangas J (2000) SOM toolbox for Matlab. Tech rep, Laboratory of Computer and Information Science, Helsinki University of Technology
Kiviluoto K (1996) Topology preservation in self-organizing maps. In: Proceedings of the IEEE international conference on neural networks, pp 294–299
Villmann T, Herrmann RDM, Martinez T (1997) Topology preservation in self-organizing feature maps: exact definition and measurement. IEEE Trans Neural Netw 8(2):256–266
Bauer H-U, Pawelzik K (1992) Quantifying the neighborhood preservation of self-organizing maps. IEEE Trans Neural Netw 3(4):570–578
Venna J, Kaski S (2001) Neighborhood preservation in nonlinear projection methods: an experimental study. In: Lecture notes in computer science, vol 2130, pp 485–491
Venna J (2007) Dimensionality reduction for visual exploration of similarity structures. Dissertation, Helsinki University of Technology
Romesburg HC (1984) Cluster analysis for researchers. Krieger, Florida
Guyon I, Elisseeff A (2003) An introduction to variable and feature selection. J Mach Learn Res 3:1157–1182
Xing EP, Karp RM (2001) Cliff: clustering of high-dimensional microarray data via iterative feature filtering using normalized cuts. Bioinformatics 17:S306–S315
Rakotomamonjy A (2003) Variable selection using SVM-based criteria. J Mach Learn Res 3:1357–1370
Perkins S, Lacker K, Theiler J (2003) Grafting: fast, incremental feature selection by gradient descent in function space. J Mach Learn Res 3:1333–1356
Reunanen J (2003) Overfitting in making comparisons between variable selection methods. J Mach Learn Res 3:1371–1382
Author information
Authors and Affiliations
Corresponding author
Appendix: Information enhancement method
Appendix: Information enhancement method
In the second data set, the voting data, we compared the results of the input information with those of one of our methods for detecting the importance of input neurons. Before this input information method, we developed two types of information-theoretic methods to detect the importance of input neurons, namely, information enhancement and loss. The information enhancement method focuses on a specific component in a neural network and then computes mutual information. When the mutual information can be increased by this focus on the component or the enhancement of the component, we considered the component to be important. On the other hand, in the information loss method, a target component is ignored in a neural network, and then mutual information is computed. If the mutual information is decreased, the component was considered to be important. At the present stage of research, information loss has been incorporated into information enhancement. We here used a new version of information enhancement, but we present it in a simplified way, focusing on the main part of the method.
The new version of information enhancement lies in the repetition of the standard information enhancement. We present the core part of information enhancement. Let us compute the output by information enhancement. The output from the output neuron, when the tth neuron is enhanced, is computed by
As is the case with the input information, the spread parameter σ is defined by using the parameter β
However, for fair comparison, we changed the spread parameter more carefully because the parameter β was too large to examine the performance. We tried to obtain the best result in terms of the topographic error. For that, the spread parameter σ is defined by another parameter, γ(>0)
The parameter γ was computed by the other parameter β
We changed the value of α from 0.005 to 1 in steps of 0.005, and the best result in terms of the topographic error was taken.
By normalizing this output, we have the firing probability
By using this probability, we have mutual information for the kth input neuron
where
By using this mutual information, we have the firing probability for input neurons
We put this firing rate p(k) into the Eq. (15) to obtain the re-estimation rule.
Rights and permissions
About this article
Cite this article
Kamimura, R. Input information maximization for improving self-organizing maps. Appl Intell 41, 421–438 (2014). https://doi.org/10.1007/s10489-014-0525-1
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10489-014-0525-1