A neural-network-like version of the above method. The training set is summarized in the form of prototypes with non-crisp labels. The network parameters (prototype locations and class labels) are learnt by optimizing an error function. The method is described in:
T. Denoeux. A neural network classifier based on Dempster-Shafer theory. IEEE transactions on Systems, Man and Cybernetics A, 30(2):131-150, 2000. pdf