I have classification task, so I need to assign label from 5 classes to object. Because classes are ordinal $[0, 1, 2, 3, 4]$ a special architecture can be used rathen then MLP with softmax.
In this case, class 0 is age from [0-10], 1 age from [10-20], 2 is age from [20-30] and so on.
For example, in this question they propose to use such ordering: $$P(0) < P(1) < P(2) < P(3)< P(4) $$
But what if I want that neural net should produce another output?
Imagine that NN produces output with $P_{arg\_max}=2$. Because class 2 is more similar to 1 and 3 I want that ordering should be like:
$$P(0) < P(1) < P(2) \text{ and } P(2) > P(3) > P(4) $$
In another words, distribution on predicted age should be unimodal.
Do you know how this can be implemented?
