本論文主要是用來改善自我組織圖，因為輸入資料與輸出的神經元陣列維度不同，所造成輸入資料的相鄰關係在輸出的神經元陣列中保存不良的情形。
自我組織圖是模仿人類大腦處理訊息的一個很簡單而又有效的方法。常應用在圖形辨識和語音辨識等領域。但是對於在輸出神經元陣列中保存輸入資料的相鄰關係，因為輸出、入維度的不同，很難有良好的結果。因此一個在學習中可以漸增神經元陣列的維度的自我組織圖，在相鄰關係保存要求較高的情況下，提供了一個解答。
本論文的作法是以有效辨別半徑來決定是否新增一個神經元，而後在新增神經元時考量是否應將輸出的神經元分佈漸增其維度，然後獲得較佳相鄰關係的保存。
根據模擬結果，本論文所提的方法，不但在保持相鄰關係上較其他方法佳，且只需要較少數目的神經元就可以達到如此結果。

The main purpose of this method is to improve the neighborhood preservation of self-organizing maps. If there’s a dimension difference of input data between output neuron arrays, it is hard to preserve the neighborhood relationship of the input data in the output neuron array.

Self-organizing maps are an easy and effective methods which imitate signal processing mechanism of humans. Usually, photographs and speech recognition are applied. However due to the dimension difference of input data and output neuron array, it’s difficult to preserve neighborhood relationships. And for some applications, higher capability of neighborhood preservation is required so a dimension-increasing self-organizing map is presented as a solution.

This method uses effective discrimination radius to determine whether a neuron need to be added or not. While adding a new neuron, we consider whether there is a need to increase output neuron dimension for preserving better neighborhood relationships.

According to simulation results, not only the neighborhood preservation is better but also a smaller number of neurons needed than other methods.

參考文獻

[1] Teuvo Kohonen,"The Self-Organizing Map," Proceedings of the IE-
EE, vol. 78, no. 9, pp. 1464-1480, Sep. 1990.
[2] Hans-Ulrich Bauer and Thomas Villmann, "Growing a Hypercubical Output Space in a Self-Organizing Feature Map," IEEE Transactions on Neural Networks, vol. 8, no. 2, pp. 218-226, Mar. 1997.
[3] Justine Blackmore and Risto Miikkulainen, "Incremental Grid Growi- ng: Encoding High-Dimensional Structure into a Two-Dimensional F- eature Map," Proceedings of IEEE International Conference on Neur- al Networks, vol. 1, pp. 450-455, Mar. 1993.
[4] Bernd Fritzke, "Unsupervised Clustering with Growing Cell Structure, "Proceedings of IJCNN-91-Seattle International Joint Conference on Neural Networks, vol. 2, pp. 531-536, Jul. 1991.
[5] Victoria J. Hodge and Jim Austin, "Hierarchical Growing Cell Struct- ures: TreeGCS," IEEE Transactions on Neural Networks, vol. 13, no. 2, pp. 207-218, Mar./Apr. 2001.
[6] Vanco Burzevski and Chilukuri K. Mohan, "Hierarchical Growing Ce- ll Structures," Proceedings of IEEE International Conference on Neu- ral Networks, vol. 3, pp. 1658-1663, Jun. 1996.
[7] Masahiro Iwasaki, Tomonori Hashiyama and Shigeru Okuma, "Self- Creating and Organizing Neural Networks with Weight Duplication," Proceedings of the IEEE-INNS-ENNS International Joint Conference on Neural Networks, vol. 4, pp. 602-607, Jul. 2000.
[8] Pengfei Xu and Chip-Hong Chang, "Self-Organizing Topological Tree, " Proceedings of the 2004 International Symposium on Circuits and Systems, vol. 5, pp. V732-V735, May 2004.
[9] Siming Lin and Jennie Si, "Dynamic Topology Representing Networ- ks," Proceedings of the 1998 IEEE International Joint Conference on Neural Networks, vol.1, pp. 353-358, May 1998.
[10] Ressom H., Wang D. and Natarajan P, "Adaptive Double Self-organ- zing Map and its Application in Clustering Gene Expression Data," Proceedings of the International Conference on Neural Networks, vol. 1, pp. 39-44, Jul. 2003.
[11] Yi Wang, Chuahua Zeng, Ting Mei and Wenqing Liu, "The Diffuse Self-Organizing Map," Proceedings of IEEE International Confere- nce on Systems, Man and Cybernetics, vol. 4, pp. 3530-3535, Oct. 2003.
[12] Kimmo Kiviluoto, " Topology Preservation in Self-Organizing Maps, " Proceedings of IEEE International Conference on Neural Networ- ks, vol. 1, pp.294-299, Jun. 1996.
[13] James S. Kirk and Jacek M. Zurada, "A Two-Stage Algorithm for I- mproved Topography Preservation in Self-Organizing Maps," Proce- edings of IEEE International Conference on Systems, Man and Cyb- ernetics, vol. 4, pp. 2527-2532, Oct. 2000.
[14] James S. Kirk and Jacek M. Zurada, "Motivation for a Genetically- Trained Topography-Preserving Map," Proceedings of the 2002 Inte- rnational Joint Conference on Neural Networks, vol. 1, pp. 394-399, May 2002.
[15] Arthur L. Hsu and Saman K. Halgamuge, "Enhanced Topology Pres- ervation of Dynamic Self-Organizing Maps for Data Visualisation," Proceedings of IFSA World Congress and 20th NAFIPS Internation- al Conference, vol. 3, pp.1786-1791, Jul. 2001.
[16] Wingho Shum, Huidong Jin and Kwongsak Leung, "A Self-Organiz- ing Map with Expanding Force for Data Clustering and Visualizati- on," Proceedings of IEEE International Conference on Data Mining, pp. 434-441, Dec. 2002.
[17] Thomas Villmann, Ralf Der, Michael Herrmann, and Thomas M. M- artinetz, "Topology Preservation in Self-Organizing Feature Maps: Exact Definition and Measurement," IEEE Transactions on Neural Networks, vol. 8, no. 2, pp. 256-266, Mar. 1997.