The Topographic Neural Gas
We have recently investigated a family of algorithms which use the underlying latent space model developed for the Generative Topographic mapping(GTM) but which train the parameters in a different manner. Our first model was the Topographic Product of Experts (ToPoE) which is fast but not so data-driven as our second model, the Harmonic Topographic Mapping (HaToM). However the HaToM is much slower to train than the ToPoE. In this paper we introduce ideas from the Neural Gas algorithm to this underlying model and show that the resulting algorithm has faster convergence while retaining the good quantization properties of the HaToM.
KeywordsData Space Vector Quantization Topology Preservation Generative Topographic Mapping Topology Preserve
Unable to display preview. Download preview PDF.
- Bishop, C.M., Svensen, M., Williams, C.K.I.: GTM: The Generative Topographic Mapping. Neural Computation (1997)Google Scholar
- Cottrell, M., Hammer, B., Hasenfu, A., Villmann, T.: Batch neural gas. In: WSOM (2005)Google Scholar
- Fritzke, F.A.: Growing Neural Gas Network Learns Topologies. In: Advances in Neural Information Processing Systems 7 (NIPS 1994), pp. 625–632. MIT Press, Cambridge (1995)Google Scholar
- Fyfe, C.: Two topographic maps for data visualization. Data Mining and Knowledge Discovery (2006)Google Scholar
- Neural Networks Research Centre, Helsinki University of Technology, SOM Toolbox, http://www.cis.hut.fi/projects/somtoolbox/
- Peña, M., Fyfe, C.: Model- and Data-driven Harmonic Topographic Maps. WSEAS Transactions on Computers 4(9), 1033–1044 (2005)Google Scholar
- Peña, M., Fyfe, C.: Outlier Identification with the Harmonic Topographic Mapping. In: 14 th European Symposium on Artificial Neural Networks, ESANN (2006)Google Scholar
- Zhang, B.: Generalized K-Harmonic Means – Boosting in Unsupervised Learning. Tech. report. HP Laboratories, Palo Alto (2000)Google Scholar