This chapter introduces a variety of learning algorithms for distinct pattern recognition tasks. The chapter consists of two sections. The first section presents training algorithms for single-layer biomimetic neural networks, while the focus of the second section is on multi-layer biomimetic neural networks. At the beginning of Section 9.1, the problem considered is the separation of two finite and nonintersecting subsets of Euclidean n-space. A well known example of this class is the XOR problem. Two distinct algorithms able to solve this problem, known as “Training based on elimination” and “Training based on merging”, are specified in the first two subsections of this chapter. Subsection 9.1.3 presents a third algorithm for multi-class single-layer biomimetic neural networks. The final subsection of Section 9.1 introduces a training method based on dual lattice metrics. The duality provides for a smaller polytope containing a data set than that of a single polytope such as an interval. The result is the reduction of errors in some pattern recognition tasks.

The focus of the second section of this chapter is on learning in multi-layer biomimetic neural networks. This necessitates both, the construction of the multilayer network and the learning phase of the network for the pattern recognition task involved. The section provides an in-depth study with examples and comparison with other other known methods.