Since data analysis tasks often have to deal with complex data structures, the nonlinear
dimensionality reduction methods such as ISOMAP and LLE need to be used in exploratory data
analysis. These techniques attempt to preserve either the local or the global geometry of the original
data, and they perform implicit mapping between the original data space and an embedding space.
Nevertheless, these methods are not compatible with large-scale datasets. There is a need to
improve existing methods by using a combination of vector quantization (VQ) and mapping methods.
VQ can be used for data representation and nonlinear mapping is used in order to visualize the
internal structure in a low-dimensional vector space. In this work, we will introduce VQ-based method,
i.e. the online visualization neural gas (OVING), for extracting the data structure and finding explicit
mapping between two data spaces. Interestingly, an explicit mapping, either forward or inverse
mapping, can be used as a substitute for supervised learning algorithm.
Thus, this mapping is useful for the tasks of system identification and pattern recognition. To
identify the benefits of OVING based mapping, several applications are demonstrated on problems of
system identification and pattern classification.