INTRODUCTION TO THE TLS EXIN NEURON - Neural-based Orthogonal Data Fitting: The EXIN Neural Networks - page 131

Figure 4.8 TLS EXIN acceleration methods for κ = 0 . 4; an epoch represents a block of
500 training vectors.
Figure 4.9 TLS EXIN acceleration methods for κ = 1; an epoch represents a block of 1200
training vectors.
in the divergence domain as seen in the analysis of convergence. TLS EXIN is
still convergent, but better results are obtained by its own acceleration techniques.
Figure 4.12 shows that even in the presence of very high noise, the BFGS and
SCG methods still yield satisfactory results.