Curvilinear Component Analysis for High-Dimensional Data Representation: II.Examples of Additional Mapping Constraints in Specific Applications

摘要

Using a recent algorithm for non linear mapping, Curvilinear Component Analysis, we show through three applicatiosn how a priori knowledge can be introduced in the CCA framework, and we translate this knowledge in term of mapping constraints. This a priori knowledge can the introduced to constraint the convergence of the algorithm toward a data structure having a best interpretation according to the physical process of input data generation. The three applications concern geographical data representation, speech recognition and IRMf image processing.