7.5. Multiclass Probabilistic DBNNs
Similar to the previous two-class (one-subnet) probabilistic DBNNs, the PDBNNs adopt a modular structure for a multiclass classification problem: one subnet is designated to one object class. The output of a subnet represents the class's likelihood function. The outputs of the subnets compete with each other, and the winner can claim the identity of the input pattern. For the multiclass probabilistic DBNNs, a mixture of Gaussians is often adopted as the class-likelihood function p(xt|Ωi).
Thus, the discriminant function and the network parameters are expressed according to a probabilistic format. For example, the discriminant function is expressed as:
Equation 7.5.1
where the i-th subnet is parameterized ...
Get Biometric Authentication: A Machine Learning Approach now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.