Biometric Authentication: A Machine Learning Approach

4.5. Nonlinear SVMs

It is well known that, by inserting a well-designed nonlinear hidden-layer between the input and output layers, a two-layer network can provide an adequate flexibility in the classification of fuzzily separable data. The original linearly nonseparable data points can be mapped to a new feature space, represented by hidden nodes such that the mapped patterns become linearly separable. This is illustrated in Figure 4.8.

Figure 4.8. Use of a kernel function to map nonlinearly separable data in the input space to a high-dimensional feature space, where the data become linearly separable.

If the hidden nodes can be expressed by ...