Designing classifiers on high-dimensional learning data sets is an important task that appears in artificial intelligence applications. Designing classifiers for high-dimensional data involves learning hierarchical neural networks combined with feature selection. Feature selection aims to omit features that are unnecessary for a given prob-lem. Feature selection in formal meurons can be achieved by minimizing convex and picewise linear (CPL) criterion functions with L1 regularization. Minimizing CPL criterion functions can be associated with computations on a finite number of vertices in the parameter space. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.